# FAQ (/docs/faq)
## Money [#money]
### Is there a free plan? [#is-there-a-free-plan]
No. What you start on is a **card-free trial**: 5,000 credits shared across the entire
workspace, not per person. When those run out you pick a plan.
### What does it cost? [#what-does-it-cost]
Per seat, per month. Annual billing is 20% off.
| Plan | Monthly | Annual |
| ---------- | ------------- | ------------- |
| Team | $29 | $23 |
| Power | $99 | $79 |
| Max | $199 | $159 |
| Enterprise | Contact sales | Contact sales |
### What is a credit? [#what-is-a-credit]
A credit is the unit AI work is measured in. A chat turn, a Routine run, a web search, a
generated document, and a voice note transcription all spend credits. Your included
allowance is **10,000 credits per seat per month on Team, 40,000 on Power, and 100,000 on
Max**, and it is **pooled across the workspace** rather than fenced off per person, so a
heavy user and a light user draw from the same pool. Everyone can see the balance on the
**Billing** tab in **Settings**; admins get the full breakdown under **Usage**. Any admin
can buy a top-up.
### Who counts as a seat? [#who-counts-as-a-seat]
Every **active** person, including the owner and every admin. Pending invitations are not
seats and are never billed.
### Who can change the plan? [#who-can-change-the-plan]
Only the **workspace owner** can start, change, or cancel a plan or update the card. Any
**admin** can buy credit top-ups and view billing. If you are an admin and the billing
controls are missing, that is why.
### My plan says "Team" but I signed up on Growth or Business. [#my-plan-says-team-but-i-signed-up-on-growth-or-business]
Those plans are retired for new customers but still bill existing ones, and a
grandfathered Growth workspace displays as Team. Your billing is unchanged. Email
**[support@talentos.so](mailto:support@talentos.so)** if the numbers look wrong.
## Privacy and people [#privacy-and-people]
### Can my manager or an admin read my chats? [#can-my-manager-or-an-admin-read-my-chats]
No. Conversations with your Talent and your personal memory are private to you, including
from admins and from the workspace owner. There is no view-as-member mode and no chat log
export.
What an admin **can** see, under **Usage** in Settings, is workspace activity: that a
member ran something, what kind of work it was, and what it cost in credits. They see that
work happened, not what you said.
### What happens when someone leaves? [#what-happens-when-someone-leaves]
An admin removes them from **Team**. Their membership goes inactive, any manager seats
they held are released, and the seat count re-syncs so billing adjusts. Their private
conversations and personal memory stay private and are not handed to anyone else.
Anything they published to the **Marketplace** or added to **company memory** stays with
the workspace.
The workspace owner is protected from removal. If the owner is leaving, contact
**[support@talentos.so](mailto:support@talentos.so)** before you remove anyone.
## Setup and connections [#setup-and-connections]
### What can I connect? [#what-can-i-connect]
Open **Connections** to connect supported email, calendar, document, drive, spreadsheet,
CRM, project management, database, and team chat accounts. You can also add messaging
channels for talking to your Talent. Admins can register custom MCP connectors for tools
that expose a compatible server. See [Connections](/docs/connections).
### How many accounts can I connect per app? [#how-many-accounts-can-i-connect-per-app]
One per app below Power. On **Power** and **Max** you can connect up to five accounts for
mail, calendar, chat, and drive apps, which is what you want if you run several client
inboxes.
### Do I have to use team chat? [#do-i-have-to-use-team-chat]
No. Web chat works on its own, and multiple messaging channels are available. Teams that
put their Talent in the messenger they already keep open do get more out of it.
### Why can the same team chat appear twice in Connections? [#why-can-the-same-team-chat-appear-twice-in-connections]
They are two different things. The personal connector lets your Talent read or post as
you. The workspace-level app lets people talk to their Talent inside that team chat.
Most teams want both.
### Is there anything special about the phone messaging channel? [#is-there-anything-special-about-the-phone-messaging-channel]
Yes, and it is worth knowing. Your Talent answers from a **shared TalentOS line** rather
workspace per day and 40 per shared line per day. If pairing does not go through, wait a
day or use another connected channel.
## A worked example [#a-worked-example]
A ten person agency starts on Team, giving the workspace 100,000 credits each month.
An admin invites two members and buys a top-up before a campaign launch. The owner later
switches the plan to Power because two strategists need several client email accounts.
The admin could view billing and buy credits, but only the owner could change the plan.
## What to check [#what-to-check]
### My Routine did not run. [#my-routine-did-not-run]
Check these in order.
1. The Routine is **enabled**, not paused.
2. It is not showing **Needs attention**, which almost always means an expired connection.
3. The **time zone** on the schedule is yours.
4. **Settings**, then **Activity**, shows a run. Every run leaves exactly one receipt. No
receipt means it never started. A receipt with a weak result means the instructions
need work, not the schedule.
5. Your workspace still has credits, on the **Billing** tab in **Settings**.
Full walkthrough: [Your first
Routine](/docs/getting-started/your-first-routine).
### My Talent says it cannot reach one of my tools. [#my-talent-says-it-cannot-reach-one-of-my-tools]
That is deliberate. It names a missing or broken connection instead of improvising an
answer. Open **Connections**, reconnect that app, and ask again.
### How do I get support? [#how-do-i-get-support]
Email **[support@talentos.so](mailto:support@talentos.so)**. Include your workspace name and, if it is about a specific
run, roughly when it happened so we can find the receipt.
# TalentOS Help (/docs)
TalentOS gives everyone on your team a personal AI teammate, called a **Talent**. Your
Talent works with the tools you connect, delivers results where you already work, and
keeps repeatable jobs moving through Routines. These guides show you how to get each job
done.
**New here?** Start with [What is TalentOS?](/docs/getting-started/what-is-talentos), then
set up [Your first Routine](/docs/getting-started/your-first-routine). One working Routine
is the fastest way to understand how the rest fits together.
## Browse by area [#browse-by-area]
## Questions people ask most [#questions-people-ask-most]
* [Is there a free plan?](/docs/faq) There is a card-free trial, not a free plan.
* [Can my manager read my chats?](/docs/faq) No.
* [My Routine did not run.](/docs/routines/fix-a-routine-that-is-not-running) Check these five
things in order.
Still stuck? Email **[support@talentos.so](mailto:support@talentos.so)**.
# Track credits and top up (/docs/billing/credits)
Credits are how TalentOS bills AI work. Every message your Talent answers and every Routine run debits the workspace pool. This page shows you where to see what is left and how to add more before anything stops.
## Where to look [#where-to-look]
Open **Settings** from the Manage group. Members go to the **Billing** tab; owners and admins go to **Plans & Billing**. Either way, the Workspace credits card shows three things that always agree with each other:
* a headline like "84,000 of 120,000 credits left"
* the percentage still spendable
* a bar that empties as the month goes on
Members see this card too. It is read-only for them, but everyone can see the same workspace truth rather than guessing.
## How the pool works [#how-the-pool-works]
* Your monthly allowance is **seats multiplied by your plan's per seat credits**: 10,000 on Team, 40,000 on Power, 100,000 on Max.
* The allowance is **pooled across the whole workspace**, not split per person. One person can use most of it in a heavy week.
* It **resets each month**. Unused included credits do not roll over.
* Purchased top-up credits sit **behind** the monthly pool and are only spent once the monthly allowance is gone. Top-ups **never expire**.
Credits are the only unit TalentOS bills or displays. Every screen, receipt and usage view is denominated in credits, so you never have to convert anything.
## What makes a task cost more [#what-makes-a-task-cost-more]
The **model lane** your Talent uses. Every lane is available on every plan, and a pricier lane simply debits more credits per task. If a workspace is burning credits faster than expected, the model lane is the first thing to check, followed by how often your Routines run.
A rough anchor on the cheapest governed lane: a tool heavy task (read a mailbox, pull a sheet, write a document) lands around 9 credits. A Team seat's 10,000 credits covers roughly a thousand of those a month.
## Buy a top-up [#buy-a-top-up]
**Any admin** can buy top-up credits. You do not need to be the workspace owner, and you do not need to change plan.
1. Open **Settings**, then **Plans & Billing**.
2. Find the Workspace credits card and pick a top-up pack.
3. Complete the purchase in the secure billing portal.
4. The credits land on the workspace balance and never expire.
## A worked example [#a-worked-example]
An eight seat Team workspace has 80,000 credits a month. Three weeks in, an ops lead sees "6,200 of 80,000 credits left" and a daily Routine that moves project updates into the team knowledge base still due to run nine more times. Rather than wait for the reset, an admin buys a top-up pack that afternoon. The Routine keeps running, and the banked credits carry into next month because they are only spent after the monthly pool.
## If credits run out [#if-credits-run-out]
Your Talents stop working until the pool is topped up or the month resets. Nothing is deleted, and everything stays readable. Two different messages mean two different things:
* **"Your workspace is out of credits"**: the shared pool is empty. An admin needs to buy a top-up, or you wait for the monthly reset.
* **A personal safety limit was reached**: one member has hit their own monthly fairness limit while the workspace pool still has credits. An owner or admin can reset that member from the usage view. See [See what your workspace spends](/docs/billing/usage-dashboard).
## What to check [#what-to-check]
* **The credits card says nothing is available.** Billing has not finished syncing, or the workspace is not on a plan yet. Check [Fix a read-only workspace](/docs/billing/read-only-mode).
* **You bought a top-up and the number did not move.** The purchase completes in the billing portal; give it a moment and reload Settings.
* **The numbers look too small for your seat count.** Confirm your seat count and plan on the same tab; a grandfathered plan keeps its original allowance. See [Older plan names](/docs/billing/older-plan-names).
# Change plan, update your card, or cancel (/docs/billing/manage-your-subscription)
Everything that shapes the subscription happens on one tab, and almost all of it is restricted to one person. This page tells you who that is and what each control does.
## Who can do what [#who-can-do-what]
| Action | Owner | Admin | Manager or member |
| ---------------------------------- | ----- | ----- | ----------------- |
| View plan, seats and credits | Yes | Yes | Yes, read only |
| Start, change or cancel a plan | Yes | No | No |
| Update the card, download invoices | Yes | No | No |
| Buy credit top-ups | Yes | Yes | No |
There is exactly one owner per workspace. If an admin clicks a billing control they do not have, they get an error rather than a change, so the panel shows them "Billing is managed by the workspace owner" instead.
## Change plan or switch to annual [#change-plan-or-switch-to-annual]
1. Sign in as the owner.
2. Open **Settings**, then **Plans & Billing**.
3. On the plan cards, toggle between **monthly** and **annual**, then choose the plan you want.
4. Confirm in the secure billing portal. The change applies to the whole workspace immediately.
Annual saves 20% against the monthly rate. Switching cadence keeps you on the same plan; switching plan changes your credits per seat and your accounts-per-app allowance. See [Choose a plan](/docs/billing/plans-and-pricing).
## Update the card, or get an invoice [#update-the-card-or-get-an-invoice]
1. Sign in as the owner.
2. Open **Settings**, then **Plans & Billing**.
3. Use the manage button on your current plan card. It opens a **secure billing portal**.
4. In the portal, update the payment method, download invoices, or cancel.
TalentOS never stores your card details. Cards, invoices and cancellation all live in that portal, which is why the button takes you out of the app.
## Seats change on their own [#seats-change-on-their-own]
You never buy seats by hand. One active person equals one seat, at the same price whatever their role. Accepting an invite adds a seat and removing someone frees it, and your invoice follows automatically. Pending invitations are never billed.
## Cancel and come back [#cancel-and-come-back]
Cancelling in the portal leaves the plan running until the end of the period you have paid for. Your plan card shows **Cancels on** with that date instead of **Next renewal**. After that date the workspace goes read-only and everything stays readable.
To come back, the owner picks a plan again on the same tab. There is no second free trial for a workspace that has already used one.
## A worked example [#a-worked-example]
A consultancy's founder is the owner and their ops manager is an admin. Mid month, the ops manager notices credits running low and buys a top-up without involving anyone, because top-ups are an admin action. At renewal the founder switches the workspace from Team monthly to Power annual so two consultants can each connect a second client email account.
## What to check [#what-to-check]
* **The controls are missing or say billing is owner managed.** You are an admin, not the owner. Ask the owner, who is marked with an Owner badge in the Team directory and on the usage view.
* **The plan card is blank right after loading.** Ownership is still resolving. Give it a second rather than assuming you lack access.
* **The portal button does nothing.** A popup blocker is the usual cause. Allow popups for the app and try again.
* **The workspace is read-only and you want to fix it.** See [Fix a read-only workspace](/docs/billing/read-only-mode).
# Older plan names (/docs/billing/older-plan-names)
If your plan is called something you cannot find on the pricing page, you are on a **grandfathered plan**. Nothing is wrong, nothing is being overcharged, and you do not have to move.
## The retired plans [#the-retired-plans]
**Growth** and **Business** are no longer sold. Workspaces that bought them before they were retired keep them, keep billing on them, and keep exactly the entitlement they were sold.
Both sit at the same allowance as **Team** today:
* **10,000 included credits per seat, per month**, pooled across the workspace.
* **One connected account per app, per person.**
Every capability is available on them, just as it is on every current plan. Nothing was taken away when the plans stopped being sold.
## Why a Growth workspace can say "Team" [#why-a-growth-workspace-can-say-team]
Early customers bought the $29 per seat tier under the name **Growth**. That same $29 tier is called **Team** today. So a grandfathered Growth workspace displays as **Team** across the app, because that is the honest current name for the plan it holds and the price it pays.
A workspace that is on a Growth price without that grandfather status still displays as **Growth**.
Either way, the price on your invoice is the price you signed up for.
## What you can change [#what-you-can-change]
* **Cadence.** A grandfathered workspace can switch between monthly and annual on its current plan.
* **Plan.** You can move up to Power or Max at any time for the larger credit pool and the higher accounts-per-app allowance. See [Choose a plan](/docs/billing/plans-and-pricing).
Moving off a retired plan is **one way**. Once you leave Growth or Business you cannot buy back onto them, because they are not sold any more.
## A worked example [#a-worked-example]
A support lead opens **Settings, Plans & Billing** and sees "Team" on the plan card, but the invoices in the billing portal say Growth. Both are right: the workspace bought Growth at $29 per seat, that tier is now called Team, and the invoice keeps the original name because the subscription itself never moved. Their credits, 10,000 per seat per month, are the Team figure, which is what they have always had.
## What to check [#what-to-check]
* **Your plan card shows a name with no matching card in the grid.** That is expected on a retired plan. Everything still works.
* **You cannot find Growth or Business in the pricing table.** They are not sold any more, by design.
* **Your credits look lower than a plan of that price should give.** Retired plans sit at the Team allowance. Compare against 10,000 credits per seat, not Power or Max.
* **You want to be certain what you are billed.** Only the owner can open the secure billing portal, where the real invoices live. See [Change plan, update your card, or cancel](/docs/billing/manage-your-subscription).
# Choose a plan (/docs/billing/plans-and-pricing)
Every plan gives you every capability. What you are choosing is **how many credits your workspace gets each month** and **how many accounts of one app each person can connect**. There is no free plan; new workspaces start on the [free trial](/docs/billing/start-a-trial).
## The plans [#the-plans]
| Plan | Per seat, monthly | Per seat, annual | Included credits per seat, per month |
| ---------- | ----------------- | ---------------- | ------------------------------------ |
| Team | $29 | $23 | 10,000 |
| Power | $99 | $79 | 40,000 |
| Max | $199 | $159 | 100,000 |
| Enterprise | Contact sales | Contact sales | Negotiated |
Annual billing is **20% off** the monthly rate, charged yearly and shown as the lower per seat monthly figure.
Included credits are **pooled across the workspace**. A ten seat Team workspace has 100,000 credits a month for everyone to share, so the person running the heavy project this month is not throttled while a quieter teammate's allowance sits idle. See [Track credits and top up](/docs/billing/credits).
## What is the same on every plan [#what-is-the-same-on-every-plan]
* A personal Talent for every member, with its own memory and connected tools.
* Routines that run on demand and on a schedule.
* Skills, Artifacts (DOCX, PDF, PPTX, XLSX, and web/HTML), Memory and the Marketplace.
* **Every model lane**, from the cheapest governed lane to the frontier ones. Picking a pricier lane costs more credits per task, not a higher plan.
* Web chat, connected messaging channels, and every live connector.
## What actually changes [#what-actually-changes]
**Credits per seat**, as in the table above.
**Accounts per app.** On **Team**, each person connects one account per app. On **Power** and **Max**, each person can connect up to **five** accounts for mail, calendar, chat, and drive tools, and up to two for every other app.
**Support and onboarding.** Power adds priority support. Max adds early access to new capabilities and a guided onboarding session. Enterprise adds annual invoicing, custom terms, security review support with a DPA, dedicated onboarding and volume credit pricing.
## How seats are counted [#how-seats-are-counted]
A seat is **one active person in the workspace**. Owners, admins, managers and members all cost the same. Pending invitations are **not** seats and are never billed. When someone accepts an invite the seat count goes up automatically; when you remove someone it goes down.
## A worked example [#a-worked-example]
A twelve person marketing team runs a Routine every weekday morning that pulls yesterday's CRM activity into a spreadsheet and posts a summary to team chat, plus ad hoc research and drafting all day. On **Team** they have 120,000 credits a month pooled. Two strategists also need a second email account for client work, which Team does not allow, so they move to **Power**: 480,000 pooled credits a month and up to five email accounts per person, at $99 per seat monthly or $79 annually.
## What to check [#what-to-check]
* **Power or Max is missing from the plan grid.** Reload once, then contact support if the card is still missing.
* **The plan you are on is called something you do not recognise.** See [Older plan names](/docs/billing/older-plan-names).
* **You want to change plan.** Only the workspace owner can. See [Change plan, update your card, or cancel](/docs/billing/manage-your-subscription).
# Fix a read-only workspace (/docs/billing/read-only-mode)
A read-only workspace has not lost anything. Every thread, Artifact, Memory page and Routine is still there and still readable. What pauses is **new work**: chat, Routines, Connections and Artifacts. This page gets you back.
## Why it happened [#why-it-happened]
The banner across the top tells you which of three states you are in.
* **"Your workspace isn't on a plan yet."** Nobody has started the trial or bought a plan.
* **"Your plan has ended."** The subscription was cancelled or the trial lapsed. The way back is a new plan, not a card update.
* **"Your plan isn't active."** A payment failed. The way back is updating the card.
## Unlock it [#unlock-it]
Only the **workspace owner** can do any of these.
1. Click the banner, or open **Settings** then the **Plans & Billing** tab.
2. Take the action that matches your state:
* Never subscribed: click **Start free trial**. See [Start your free trial](/docs/billing/start-a-trial).
* Plan ended: choose a plan from the cards.
* Payment failed: open the secure billing portal from the manage button and update the card.
3. Wait a few seconds. The banner clears by itself and the workspace unlocks. You do not need to sign out or ask anyone to reload.
## What everyone else should do [#what-everyone-else-should-do]
Nothing. Admins, managers and members see the same banner with no action attached, because billing writes belong to the owner alone. They can keep reading everything while they wait. If you are an admin and the banner offers you no button, that is expected rather than a bug.
Scheduled Routines do not run while the workspace is read-only, and they do not backfill the runs they missed once you unlock. If a Routine had to produce something during the outage, ask your Talent for it directly afterwards.
## A worked example [#a-worked-example]
An agency's card expires over a holiday. On Monday everyone sees "Your Team plan isn't active, so the workspace is read-only", and the daily standup Routine has not posted to team chat. Two admins open billing and get a read-only view with a note that billing is owner managed. The founder, who is the owner, opens **Settings, Plans & Billing**, enters the new card through the secure billing portal, and the workspace becomes writable again. They then ask their Talent to produce Monday's standup summary by hand.
## What to check [#what-to-check]
* **You updated the card and it is still read-only.** Billing sync takes a moment. Reload once. If it persists past a couple of minutes, contact support.
* **You are the owner but the button is missing.** Ownership can still be resolving right after page load. Give it a second before assuming a permissions problem.
* **The banner says the plan ended but you expected a card prompt.** A cancelled subscription cannot be revived with a card. Pick a plan instead.
* **You have credits left but everything is still stopped.** Credits and plan status are separate. A workspace with credits but no active plan is still read-only. Running out of credits, by contrast, is not read-only mode. See [Track credits and top up](/docs/billing/credits).
* **You cannot reach the owner.** Contact support with your workspace name.
# Start your free trial (/docs/billing/start-a-trial)
If a new workspace is still read-only after setup, the owner can start its card-free trial from Billing. The trial includes the full self-serve product and a shared credit grant.
## Start it [#start-it]
1. Sign in as the **workspace owner**. Only the owner can start a plan or a trial.
2. Open **Settings** from the Manage group in the left nav, then the **Plans & Billing** tab.
3. Click **Start free trial**.
4. Wait a few seconds for billing to sync. The read-only banner clears on its own, and chat, Routines, Connections and Artifacts unlock.
If someone else on your team clicks it, they get an error. Send them to the owner instead.
## What the trial includes [#what-the-trial-includes]
* **Every self-serve capability.** Nothing is held back for a paid tier. Your Talent, Routines, Skills, Artifacts, Memory, the Marketplace and all model lanes are available during the trial exactly as they are on a paid plan.
* **5,000 credits, shared.** The trial grant belongs to the **whole workspace**, not to each person. If you invite ten teammates, they all draw from the same 5,000 credits. Your Billing tab shows how many are left and when the trial ends.
* **No card.** You are not asked for a payment method to begin, and nothing is charged when the trial ends.
The trial grant is shared across the workspace, so inviting your whole team on day one will burn through it fast. Try it with two or three people first, then invite the rest once you have picked a plan.
## One trial per workspace [#one-trial-per-workspace]
Each workspace gets exactly one trial. If your trial has already been used, the button is replaced by the plan cards and you choose a paid plan instead. Creating a second workspace to get a second trial is not something we support, and it splits your Memory, Connections and Marketplace across two places that cannot see each other.
## A worked example [#a-worked-example]
A six person agency signs up. The owner starts the trial and connects email and team chat. Two account managers each ask their Talent to summarise the week's client mail into a document. Between them they spend around 400 credits. A few days later the owner opens **Settings, Plans & Billing**, sees roughly 4,600 credits left and the trial end date, and buys Team for six seats before the trial lapses.
## When the trial ends [#when-the-trial-ends]
The workspace goes read-only. Nothing is deleted, everyone can still read every thread, Artifact and Memory page, but new work stops until the owner picks a plan. See [Fix a read-only workspace](/docs/billing/read-only-mode).
## What to check [#what-to-check]
* **The button is missing.** You are not the owner. Any admin can see the Billing tab, but only the owner can act on it.
* **You get "already used its free trial".** This workspace has trialled before. Choose a paid plan on the same tab.
* **The banner is still there after clicking.** Billing sync takes a moment. Reload the page once. If it persists for more than a minute or two, contact support.
* **Credits ran out before the days did.** The trial ends whichever comes first. Pick a plan, or ask support about a top-up.
## Next [#next]
# See what your workspace spends (/docs/billing/usage-dashboard)
When the credit pool drains faster than you expected, the Usage tab tells you exactly where it went over the last 30 days, in credits, per member and per Routine.
## Open it [#open-it]
1. Sign in as an **admin** or the **owner**.
2. Open **Settings** from the Manage group.
3. Choose the **Usage** tab.
Everything on this tab covers a rolling **30 day window** and is denominated in credits.
## The four views [#the-four-views]
**Overview.** Total spend for the window, your burn rate, and where it goes. Underneath, Top members and Top routines rank the biggest consumers so you can start with the one that matters.
**Team.** A sortable table of every member: conversations, Routine runs, credits spent, and last activity. Sort by credits to find your heaviest user in one click.
**Activity.** A spend ledger, newest first: one row per conversation turn or Routine run, what kind of work it was, which model lane ran it, and what it cost. Filter it to a single member when you need to explain one person's number. It records that work happened and what it charged, never what was said.
**Routines.** What each scheduled Routine costs to keep running, with how often it fires and its average cost per run. A Routine that runs every weekday is the most common reason a bill moves without anyone noticing.
## Reading it correctly [#reading-it-correctly]
* **People who have left are still listed.** Their spend was real and happened inside the window. Dropping them would stop the column adding up to the workspace total. They carry a Removed badge.
* **A `≥` in front of a number means "at least".** Very busy workspaces have more 30 day activity than one bounded query returns, so those values are lower bounds, and a note under the table says so.
* **Private Routines show as "Private routine".** If a teammate's Routine is private, you see its cost but not its title. That is deliberate: spend is workspace business, the contents of someone's private work are not.
## Reset a member's personal safety limit [#reset-a-members-personal-safety-limit]
Alongside the workspace pool, each member has a personal monthly fairness limit so one person cannot drain the shared pool in an afternoon. When someone hits theirs, their Talent stops for them while everyone else keeps working.
1. Open **Settings**, then **Usage**, then the **Team** view.
2. Find the member's row and click **Reset personal safety limit**.
3. Confirm. Their full personal limit is restored immediately.
This does not add credits to the workspace. If the shared pool is empty, resetting a person changes nothing; buy a top-up instead. See [Track credits and top up](/docs/billing/credits).
## A worked example [#a-worked-example]
An agency's Team workspace burns 60% of its monthly credits in nine days. The owner opens Usage. Overview shows Top routines led by a daily competitor-research Routine. The Routines view confirms it is running seven days a week on an expensive model lane and costing more than everything else combined. They move it to weekdays only and switch its lane down. Burn drops by roughly half, and no one has to change plan.
## What to check [#what-to-check]
* **There is no Usage tab.** You are signed in as a member. Members see the workspace credit total on the Billing tab instead. If you are an admin, reload once and contact support if it is still missing.
* **The tab says usage is unavailable.** Reload once. If it persists, contact support.
* **The numbers do not match your invoice.** Usage is a rolling 30 day window, not your billing period, and it excludes purchased top-ups still sitting in the balance.
* **A member's Routines are missing titles.** They are private. Ask the person, or read their run receipts with them.
# Connect Slack, Telegram, or iMessage (/docs/connections/connect-a-channel)
Channels are where you talk to your Talent and receive results when you are away from the web console. Slack, Telegram, and iMessage are available today.
## Connect a channel [#connect-a-channel]
1. Open **Connections** and find **Where you chat**.
2. For **Slack**, ask an admin to install the workspace app if needed, then choose **Open** or **Connect** and finish the personal chat link.
3. For **Telegram**, choose **Connect** and complete the pairing flow.
4. For **iMessage**, choose **Connect**, use the current code or phone shortcut, and send the pairing text from the phone number you want to bind.
5. Send `What can you help me with today?` as a test. Your Talent should answer in the same channel.
iMessage uses a shared TalentOS line and is available to customers on every plan, including the card-free trial. Pairing allows up to 40 new pairings per line each day and up to 10 per workspace each day. Group chats and multi-person conversations are not supported.
Voice notes work on Slack, Telegram, and iMessage. See [Talk to your Talent with voice notes](/docs/talent/voice-notes).
## A worked example [#a-worked-example]
A six person operations team wants weekday Routine results on their phones. Each person connects Telegram and sends a test message. The Routine owner then chooses Telegram under **Result delivery**, runs the Routine once on demand, and checks that the result arrives in the same private chat before relying on the schedule.
## What to check [#what-to-check]
* **A Routine ran but no channel message arrived.** Open the Routine receipt in **Activity**, then check that the intended channel is selected under **Result delivery**.
* **iMessage says Temporarily unavailable.** Keep an existing pairing if you have one and try setup again later. Do not repeatedly create new pairing codes.
* **iMessage pairing asks you to try tomorrow.** The shared line or workspace reached its daily pairing limit.
* **Slack chat works but Slack reading does not.** The workspace app and the personal Slack connector are separate. Follow [Set up both Slack connections](/docs/connections/slack-workspace-app-and-personal-connector).
* **A voice note did not send.** Send a short text in the same channel. If text also fails, reconnect the channel first.
# Connect a work app (/docs/connections/connect-a-work-app)
Connect the accounts your Talent needs for real work. Each connection belongs to you and uses the access you already have in that app.
## Apps available today [#apps-available-today]
You can connect these work apps now:
| Work area | Apps |
| ------------------ | ---------------------------------------------------------- |
| Email and calendar | Gmail, Google Calendar, Outlook |
| Documents and data | Airtable, Google Docs, Google Drive, Google Sheets, Notion |
| Project work | Asana, Jira, Linear |
| Sales | HubSpot |
| Development | GitHub |
| Communication | Slack personal connector |
**Klaviyo, Fireflies, and X are coming soon.** Their cards do not have a working Connect action and they are not connectable today.
## Connect one account [#connect-one-account]
1. Open **Connections**.
2. Search for the app or narrow the catalog by category.
3. Open the app card and choose **Connect**.
4. Sign in to the account you want your Talent to use and approve the requested access.
5. Return to Connections and wait for the card to show **Connected**.
6. Ask your Talent to do one small read, such as finding a document or listing today's events.
Your Talent sees only what that connected account can already reach. Connecting an app does not change the account's permissions.
## Connect more than one account [#connect-more-than-one-account]
Plans below Power include one account per app. Power and Max can connect up to five accounts for eligible mail, calendar, chat, and drive apps.
Open a connected app and choose **Add account** when the option is available. Give each account a clear label, such as `Client inbox` or `Company calendar`, so you can name the right source in a request. Your Talent will not guess between accounts when the choice matters.
## A worked example [#a-worked-example]
An agency operator handles a company inbox and a client inbox. On Power, they connect both Gmail accounts and label them `Company` and `Northstar client`. They ask:
> Find unread renewal messages in the Northstar client inbox from the last seven days. Do not use the Company inbox.
The account label makes the source explicit. Before scheduling the request as a Routine, the operator runs it once and checks that every result came from the client inbox.
## What to check [#what-to-check]
* **The card never reaches Connected.** Return to the app card and retry. If it says the sign-in expired or needs attention, reconnect the same account.
* **Add account is missing.** Your plan includes one account for that app, or that app does not support the higher multi-account allowance.
* **Your Talent asks which account to use.** Name the account label in the request or Routine instructions.
* **The app shows Coming soon.** It cannot be connected yet. Do not build a Routine that depends on it.
* **A connected account is not included after a plan change.** Open the app card to see which account remains active. Upgrade again or disconnect an unused account before adding another.
# Give your Talent access to the right tools (/docs/connections)
Connections is where you decide what your Talent can reach and where you can talk to it. Every member connects their own accounts. An admin connecting an app does not silently give everyone else access to that account.
## Choose what you are connecting [#choose-what-you-are-connecting]
* **Work apps** let your Talent read information and take governed actions in the tools you already use.
* **Channels** let you talk to your Talent outside the web console and receive Routine results or nudges.
* **Custom connections** let an owner add a supported private tool that is not in the catalog.
Slack has two separate connections. The workspace app is the channel where people chat with their Talents. The personal Slack connector lets one person's Talent read and search Slack with that person's account. Setting up one does not set up the other.
## A worked example [#a-worked-example]
Jordan wants a weekday campaign briefing delivered in Slack. Jordan connects Gmail and Google Calendar as work apps. An admin installs the TalentOS workspace app for Slack, then Jordan opens Slack and finishes the personal chat link. Jordan's Talent can now read Jordan's mail and calendar, run the Routine, and deliver the result to Jordan in Slack. If the Routine also needs to search Slack history, Jordan connects the personal Slack connector too.
## Start here [#start-here]
## What to check [#what-to-check]
* **Your Talent says an app is missing.** Open **Connections** under your own account. A teammate's personal connection does not replace yours.
* **A card says Coming soon.** It is visible for planning only and cannot be connected yet.
* **A connected app stopped working.** Open its card and reconnect it. A password, permission, or account change can expire access.
* **A write waits for you.** The connection is working. Your permission setting requires approval for that action.
# Set up both Slack connections (/docs/connections/slack-workspace-app-and-personal-connector)
Slack appears twice in Connections because it does two different jobs. Treat them as separate setup steps.
## 1. Workspace app: talk to your Talent [#1-workspace-app-talk-to-your-talent]
The TalentOS workspace app is the channel connection. It lets people direct message their Talent, mention it in a channel, receive results, and respond to approvals.
An admin installs the app once for the workspace. After that, each member chooses **Open** or **Connect** on the Slack channel row, sends their Talent a message, and follows the one-tap link if asked.
Installing the workspace app does not let your Talent search Slack history with your account.
## 2. Personal connector: use Slack as a work tool [#2-personal-connector-use-slack-as-a-work-tool]
The Slack card in the work app catalog is your personal connector. Connect it when you want your Talent to read or search channels your Slack account can access.
Each person connects this for themselves. It does not install the workspace app and it does not make a teammate's Slack access available to you.
| You want to... | You need... |
| --------------------------------------------- | ------------------------------------------ |
| Direct message your Talent | Workspace app plus your personal chat link |
| Mention your Talent in a channel | Workspace app plus your personal chat link |
| Receive a Routine result in Slack | Workspace app plus your personal chat link |
| Search Slack messages or read channel history | Personal Slack connector |
## A worked example [#a-worked-example]
Priya can direct message her Talent in Slack, but this request fails:
> Search the launch channel for every open decision from this week.
The workspace app is working because the conversation reached her Talent. The missing piece is the personal Slack connector. Priya opens **Connections**, finds the Slack work app card, connects her own account, then retries the search.
## What to check [#what-to-check]
* **You can chat, but search fails.** Connect the personal Slack connector.
* **The personal connector says Connected, but you cannot chat.** Ask an admin to install or repair the workspace app, then finish your personal chat link.
* **The workspace row says Ask an admin to install.** Only an admin can complete that workspace-level step.
* **Slack says the wrong workspace is installed.** An admin should contact support from the workspace app row rather than reinstalling at random.
* **A channel answer is brief.** Shared channels use stricter privacy rules. Ask in your direct message when the answer needs private context.
# Add your own MCP server as a connector (/docs/developers/add-your-own-mcp-server)
The built-in connectors cover the apps most teams live in. When the tool you
need is not one of them but publishes an MCP server, register that server
yourself and your Talent gains its tools on the next turn. This is the opposite
direction from [connecting an MCP client](/docs/developers/connect-an-mcp-client):
here TalentOS is the client.
## Add a server [#add-a-server]
1. Open **Connections** in the console and scroll to **Custom MCP servers**.
2. Select **Add MCP server**, then choose **MCP server** in the dialog.
3. Paste the server's HTTPS URL, for example
`https://tools.example.com/mcp`.
4. Give it a label, or leave it blank and TalentOS names it after the host.
5. Select **Connect**. The provider opens its own sign-in and consent screen in a
new tab. Approve there, and you land back on **Connections**.
The row then shows **Connected**, the scope, and how many tools the server
offers. Select **Check** at any time to re-test it.
Where you add the server sets who can use it. Added from your own Connections
page it is **Personal**, reachable only by your Talent. Added from the admin
Connections view it is **Workspace**, reachable by every member's Talent, and
that takes an admin. You can remove your own personal servers; only an admin
removes a workspace one.
## How your Talent uses it [#how-your-talent-uses-it]
Your Talent asks the server what tools it has before it uses any of them, so it
always works from the live list rather than a remembered one. Then, for every
single tool call:
1. Your Talent parks the call and asks you to approve it.
2. You see the exact server, tool, and arguments it wants to send.
3. Nothing runs until you approve, and the approval is bound to the arguments
you actually saw.
Custom MCP tool calls are always ask, and there is no setting that makes them
automatic. That is deliberate: approving "any tool on any server I ever add" is
not a decision anyone can make in advance. If you want a tool to run on a
schedule without you watching, use a built-in connector instead. See
[Connections](/docs/connections).
Each approved call runs at most once. If it fails partway, your Talent reports
the failure rather than retrying, because a remote tool may not be safe to
repeat. Approve a fresh call to try again.
## Limits worth knowing [#limits-worth-knowing]
* 10 workspace servers, and 10 personal servers per person.
* Up to 200 tools are read from any one server.
* The URL must be a public HTTPS address. Private and local addresses are
refused, including after a redirect.
* Removing a server deletes it along with the stored credentials. There is no
archive to restore from, so re-add it if you need it back.
## A worked example [#a-worked-example]
An operations team uses an internal order system that exposes an MCP server.
An admin adds its public HTTPS endpoint as a workspace server and completes the
consent screen. A member then asks their Talent to look up order 4812. The
Talent shows the exact `find_order` call and order number, waits for approval,
then returns the result. The Activity receipt records the approved call.
## What to check [#what-to-check]
Failures show as a short reason on the server's row. What each one means:
* **"That URL is not a public HTTPS address."** You pasted an `http://` URL, a
local address, or something that redirects to one. Use the provider's public
HTTPS endpoint.
* **"The server did not answer as an MCP server."** The URL is reachable but is
not an MCP endpoint. Most providers document a path ending in `/mcp`.
* **"The server rejected the access token."** The sign-in expired or was revoked
on the provider's side. Remove the row and add it again to sign in fresh.
* **"The server could not be reached."** The provider is down or slow. Select
**Check** in a few minutes.
* **No Custom MCP servers section at all.** Check that you opened the admin
Connections view for a workspace server. If you are already an admin, reload
once and contact support.
* **The section is there but your Talent says it has no such tool.** Select
**Check** on the row first. If it reports tools and your Talent still cannot
see them, start a new message so it reads the connection list again.
# Create an API key for an external tool (/docs/developers/api-keys)
You want a tool outside TalentOS, usually an MCP client or a script, to
read what your workspace has published. An API key is how you hand it that
access. This page covers making one, what it can reach, and turning it off.
Any admin can do this. Members do not see the panel.
## Create a key [#create-a-key]
1. Open **Settings** in the console and choose the **API & MCP** tab. (An older
`/admin/api-keys` link redirects here.)
2. Select **Create key**.
3. Name it after the tool that will hold it, for example `Ops reporting client`.
Up to 100 characters.
4. Select **Create key**. The full key appears once and starts with `tos_`.
5. Paste it straight into the tool, then close the dialog.
The full key is shown once, at creation. TalentOS keeps only a secured hash of
it, so nobody, including support, can read it back to you. If you lose it,
revoke that key and create another.
## What a key can and cannot reach [#what-a-key-can-and-cannot-reach]
A key is workspace-wide and unscoped. There is no way to limit one to a single
Skill, a single connector, or a single person's chat, so treat it like a
password for everything your workspace has published.
* It **reads** the work your workspace has published, plus any drafts belonging
to the person who created the key.
* It **never publishes** and never changes anything already published. A key
cannot edit shared work, cannot touch another member's private work, and
cannot reach another workspace.
* Because reads follow the person who created the key, a key made by someone
with narrow visibility sees less. Create it as the person whose view the tool
should have.
* Each key is capped at 60 requests per minute.
Most tools reach the workspace through the TalentOS MCP server, which already
knows the address to call. See
[Connect an MCP client](/docs/developers/connect-an-mcp-client).
If you need to call the HTTP endpoints directly instead, write to
[support@talentos.so](mailto:support@talentos.so) for the base address and the current endpoint list.
## Rotate or revoke a key [#rotate-or-revoke-a-key]
The list shows every key with its name, the first eight characters of the key,
when it was created, when it was last used, and whether it is **Active** or
**Revoked**.
To rotate: create the replacement key first, paste it into the tool, confirm the
tool still works, then revoke the old one. To retire a key outright, select
**Revoke** on its row and confirm.
Revoking takes effect immediately and cannot be undone. Revoked keys stay in the
list, greyed out, so you keep the record of what existed.
A workspace can hold 100 keys. Revoke the ones you no longer use rather than
collecting them.
## A worked example [#a-worked-example]
An operations lead creates a key named `Weekly reporting client`, pastes it into
the client's environment, and confirms the client can list the workspace's
published Skills. They then create a replacement key, update the client, test
one read, and revoke the old key. The revoked key stays listed for the audit
record but stops working immediately.
## What to check [#what-to-check]
* **401 Unauthorized.** The same 401 is returned for a missing, malformed,
unknown, or revoked key, on purpose. Check the header reads
`Authorization: Bearer tos_...`, that you pasted the whole key, and that the
key still shows **Active**.
* **429 Too Many Requests.** You crossed 60 requests a minute on that key. The
response carries a `Retry-After` value in seconds. Wait it out, or give the
noisy tool its own key so it cannot starve the others.
* **"Last used" still says Never.** That stamp is refreshed at most once an
hour, so a key that just worked can still read Never for a while.
* **No API & MCP tab.** You are signed in as a member. Ask an admin.
# Connect an MCP client (/docs/developers/connect-an-mcp-client)
TalentOS runs an MCP server. Connect a client to it and your coding agent can
read what your workspace has published, without you pasting anything in by hand.
Setup takes about two minutes.
You need an API key first. See
[Create an API key for an external tool](/docs/developers/api-keys).
## Connect the client [#connect-the-client]
1. Open **Settings** in the console, choose the **API & MCP** tab, and find the
**Other MCP clients** card.
2. Open the **MCP config** tab and copy the JSON block for a
`.mcp.json` file or a desktop client's config:
```json
{
"mcpServers": {
"talentos": {
"command": "npx",
"args": ["-y", "talentos"],
"env": {
"TALENTOS_API_KEY": "tos_your_key_here"
}
}
}
}
```
3. Replace `tos_your_key_here` with your real key.
4. Save the client configuration and restart the client.
5. Ask it to list your TalentOS workflows.
The published `talentos` package already knows which address to call, so the
key is the only workspace credential you supply.
## A worked example [#a-worked-example]
Your ops lead keeps a published Skill called "Weekly pipeline report". After
connecting an MCP client, ask:
> Using the talentos tools, list what our workspace has published, then show me
> "Weekly pipeline report" as a markdown runbook.
The client calls `list_workflows`, then `get_workflow`, and prints the runbook.
You can now work against it locally without copying anything out of the console
by hand.
## What the connection can and cannot do [#what-the-connection-can-and-cannot-do]
The server exposes six tools: `get_context`, `list_workflows`, `get_workflow`,
`list_nodes`, `create_skill`, and `update_skill`. The names still say "workflow"
in places where the console now says Skill or Routine.
* It acts as the member who created the key, so it sees exactly what that person
sees: everything published in the workspace, plus that person's own drafts.
* It can draft a private Skill for that person, and replace one it drafted
before.
* It never publishes, never edits published work, and never reaches another
member's private work or another workspace.
There is no browser sign-in flow for chat apps today, so a chat assistant
cannot connect itself to TalentOS by asking you to log in. Every client
connects with a workspace API key, and you can cut it off by revoking that key.
To give your Talent tools from a server you host or subscribe to, that is the
other direction. See
[Add your own MCP server as a connector](/docs/developers/add-your-own-mcp-server).
## What to check [#what-to-check]
* **The client shows no talentos tools.** Restart it. Most clients read their
MCP config only at launch.
* **`npx: command not found`.** The launch command needs Node.js on the machine
running the client. Install Node, then run the setup command again.
* **Every call returns 401.** The placeholder key is probably still in place, or
the key was revoked. Check the key shows **Active** under **Settings** and
reads `tos_...` in full.
* **Calls start failing after a burst.** Each key allows 60 requests a minute.
Pause, then retry after the `Retry-After` window.
* **Reads work but drafting a Skill fails.** A workspace whose plan has lapsed
is read-only until billing is back in good standing. Reads keep working.
* **A workflow you expected is missing.** It is probably still a draft owned by
someone else. Drafts stay private to their author, including over MCP.
# Turn a chat answer into a real file (/docs/artifacts/create-an-artifact)
An artifact is a piece of work your Talent saved instead of just typing into the chat. It is a real, durable thing: it has a title, it keeps every version, it lives in your workspace, and you can download it as a file. Chat text scrolls away. An artifact does not.
Artifacts start owner-private. Nobody else in your workspace can open one until you decide to share it.
## Ask for one [#ask-for-one]
You do not need a special button. Say what you want and say you want it saved.
1. Open **Home** and talk to your Talent, or send the same request from Slack, Telegram or iMessage.
2. Ask for the work and name the shape you want. For example: *"Pull the deals we closed last month from HubSpot, write a one page win/loss summary, and save it as a document."*
3. When the Talent finishes, a card appears under its message with the artifact's title and version.
4. Choose **Open** on that card. The artifact opens beside your chat so you can read it without losing the conversation.
If you also want a downloadable file, say so in the same breath: *"...save it as a document and render it as a PDF."* You get one artifact with a PDF you can download, not two separate things.
## A worked example [#a-worked-example]
> "Every Monday I have to tell the team what changed in the Notion roadmap."
Ask: *"Read the Q3 Roadmap page in Notion, list what moved since last Monday, and save it as a document called Roadmap Changes. Render it as a DOCX so I can paste it into the deck."*
You get one artifact named Roadmap Changes, version 1, with a DOCX in its Download menu. Next Monday, ask your Talent to update the same document. It writes version 2. Version 1 is still there.
## Where artifacts show up [#where-artifacts-show-up]
* **In chat**, as a card under the message that produced them.
* **On the Artifacts page**, under **Workspace** in the left navigation. That is the full list of everything you own.
* **In Activity**, inside Settings, as a receipt. Every run leaves exactly one receipt, and the receipt links the exact version that run produced.
## Editing [#editing]
Open the artifact and choose **Edit** in the header. A document has two views: **Preview**, which is the deliverable itself, and **Source**, which is the exact text behind it. Save exists only while you are in Source. Every save creates a new version, so you can always go back.
You can also just ask. *"Cut the last section and add a line about pricing"* revises the same artifact and bumps the version.
## What to check [#what-to-check]
* **No card appeared.** The Talent answered in prose instead of saving. Say "save that as a document" and it will write one.
* **The card says queued or running.** Rendering a file is real background work. It finishes on its own. Nothing is lost if you close the tab.
* **The card says failed.** Choose **Retry** on the card. A retry starts a fresh attempt.
* **You cannot find it later.** Artifacts are listed by title. If you did not name it, ask your Talent what it called the file, or sort the Artifacts page by **Updated**.
# Find an artifact and go back a version (/docs/artifacts/find-and-versions)
Everything your Talent saved for you lives on one page: **Workspace** then **Artifacts** in the left navigation. Every artifact keeps every version, so nothing you had last week is gone because something got rewritten today.
## Find it [#find-it]
1. Open **Artifacts**.
2. Use the scope row at the top. **Yours** is everything you own. **Shared with you** is only what your department shared with you, so it never repeats your own work. If you are not in a department, you only see **Yours**.
3. Search by name in the search box, or narrow with the **Visibility** filter.
4. Sort by **Updated** or **Created**.
Each row tells you what it actually is. A rendered document reads `Deck · PPTX`, `Spreadsheet · XLSX`, `Doc · DOCX` or `PDF`. Something with no render yet stays a plain document. A single page app carries a **Web app** badge, and anything published for coworkers to reuse says **In the Marketplace**.
## Open an exact version [#open-an-exact-version]
Open the artifact and look at the header. Next to the title you get a quiet `v3 · History` control.
1. Choose **History**.
2. Pick the version you want. It opens read only, exactly as it was saved.
3. To make it current again, choose **Restore**. That writes a new version whose content is the old one, so the versions in between still exist.
Links you copy always point at an exact version, not "whatever this is now". That is deliberate: a coworker opening the link a month later sees the words you sent them, not a later draft.
A version is written on every save, including one where you only changed the title. If two tabs try to save the same artifact, you get a conflict to resolve rather than a silent overwrite of your draft.
## Archive or delete [#archive-or-delete]
These are not the same thing, and the difference matters.
* **Archive** takes an artifact out of your lists and makes it read only. You can still open its versions, so old receipts in Activity keep working.
* **Delete** takes it out of your lists and takes the content with it. Source, versions, rendered files and previews all stop being readable. Old links, chat cards and Activity entries show `This artifact was deleted` with nothing to open, download or adopt.
Both are on the overflow menu of the card or row, and in the artifact's own header. A delete confirmation names the artifact before it goes.
## A worked example [#a-worked-example]
You sent version 3 of a launch brief to leadership, then rewrote the positioning in version 4. Open **History**, select version 3, and use its exact link when someone asks what they reviewed. If version 3 should become the starting point again, restore it. TalentOS creates a new current version and keeps both later and earlier versions intact.
## What to check [#what-to-check]
* **Something you expected is not in the list.** Check the scope row. Work shared with your department is under **Shared with you**, not **Yours**. Archived and deleted artifacts are in neither.
* **Shared with you is empty.** Nothing has been shared with your active department yet. If you belong to more than one, only the active one counts.
* **The list looks short.** Older work paginates. Scroll to load more rather than assuming it was lost.
* **A row has no format badge.** On a busy page, format checks are capped for speed. The badge is missing, the file is not.
* **You are an admin looking for a coworker's file.** Admins see a metadata inventory of the workspace's artifacts so they can rename or delete, but never the contents. Being an admin does not open someone else's work.
# Ask for a deck, spreadsheet, document or PDF (/docs/artifacts/formats)
Your Talent can turn an artifact into a real Office file. Four formats ship today:
| You ask for | You get | Shows in the list as |
| ------------------------- | ------- | -------------------- |
| a deck or slides | PPTX | Deck · PPTX |
| a spreadsheet or workbook | XLSX | Spreadsheet · XLSX |
| a Word document | DOCX | Doc · DOCX |
| a PDF | PDF | PDF |
There is a fifth kind that is not a file: a **web app**, a single self-contained page your Talent builds and runs inside a sealed preview. Use it for a dashboard or a one page site you want to look at rather than email.
## Ask for a format [#ask-for-a-format]
Name the format in the request. *"Build me an eight slide deck on Q3 pipeline from the HubSpot data and render it as PPTX."* You can also add a format to an artifact you already have.
1. Open the artifact from the **Artifacts** page or from its chat card.
2. Open the **Download** menu in the header.
3. Choose a format under **Render as...**.
4. Watch the status above the preview. It reports queued, running, or failed with a **Retry**.
**Render as...** only appears for the owner, and only on the current saved version. That is the only version a render can run from. If you want an older version as a file, restore it first, then render.
## Download it [#download-it]
The **Download** menu lists every completed format for the version you are looking at, primary format first, with **Markdown source (.md)** always last. A queued or failed render is never offered, and neither is a render that belongs to a different version, so what the menu says you can download is what you actually get.
The preview you see on screen is the file itself, not an approximation. A PDF renders as a PDF, a spreadsheet as a read only grid with real formula results, a deck as its slides in order.
## Branding [#branding]
If an admin has set a company brand, rendered files pick it up automatically: your colors, your heading and body fonts, your table and chart styling, and an approved logo. You do not choose a template.
Two honest limits worth knowing:
* Fonts come from a fixed list. Inter and Source Sans 3 ship today. A brand that names something else fails validation instead of quietly substituting a different face.
* The brand is frozen at the moment the render starts. If your admin changes the company default afterwards, files already rendered keep the look they were built with. Re render to pick up the new one.
## A worked example [#a-worked-example]
You have a campaign results artifact with the final numbers and commentary. Open it, render the current version as XLSX for the operations team and PPTX for the review meeting, then download both from the same version. Check the version number before sending so the spreadsheet and deck tell the same story.
## What to check [#what-to-check]
* **The format you want is not in the menu.** It has not been rendered yet. Use **Render as...** to start it.
* **The render fails repeatedly.** File rendering runs on a background worker. If that worker is unhealthy, a format can be temporarily unavailable for everyone. Wait a few minutes and retry; if it keeps failing, contact support.
* **A deck says its slides cannot be laid out.** That version's source text is not deck source. The PPTX download still works; the on screen slide view does not.
* **Preview says the file cannot be verified.** The stored file no longer matches what the run committed. Nothing was deleted and nothing was unshared. Re render the version to get a clean file.
* **Double clicking a format did nothing extra.** Two clicks collapse into one queued job on purpose. You are not being charged twice.
# Share an artifact with the right people (/docs/artifacts/share-an-artifact)
Artifacts are private to you until you say otherwise. There are exactly two ways to let someone else in, they do different jobs, and doing one never does the other.
Open the artifact and choose **Share** in the header. The dialog has both.
## Direct access: let your department open this one [#direct-access-let-your-department-open-this-one]
Use this when specific colleagues need to read this exact piece of work.
1. In the **Direct access** section, choose your department instead of **Private**.
2. Choose **Copy link**.
3. Paste the link wherever your team is: Slack, Telegram, email.
What the recipient gets is deliberately narrow. They sign in, they open the exact version you linked, and they read it. No editor, no history, no restore, no archive, no render, no sharing controls of their own.
Copying the link never widens access. If you paste a link before changing the audience, the person clicking it gets nothing. Change the audience first, then copy.
Direct access offers **Private** and your active department. Company wide direct access is not an option, and there is no public link. Sharing outside your workspace is not something an artifact can do; anyone opening the link must be a signed in member of your workspace.
## Marketplace: let the whole company reuse it [#marketplace-let-the-whole-company-reuse-it]
Use this when the artifact is a template other people should start from, not a document two people need to read.
1. In the **Marketplace** section of the same dialog, list the current version.
2. Coworkers now find it under **Marketplace** and add their own private copy.
3. When you improve the artifact, come back and choose **Update listing to v4** (or whatever version you are on). The listing is frozen at the version you published, so editing the artifact does not silently change what everyone else sees.
4. **Remove** takes the listing down. Your artifact is untouched.
A listed artifact shows **In the Marketplace** on its row so you can tell at a glance.
## The two are independent [#the-two-are-independent]
* Publishing to the Marketplace does not give anyone direct access to your live artifact.
* Giving your department direct access does not put anything in the Marketplace.
* Whatever you do, ownership stays yours.
## Web apps are different [#web-apps-are-different]
A web app can be given department direct access and opened read only through its link. It never goes to the Marketplace and it is never public. If you download the HTML and open it outside TalentOS, it loses the sandbox that keeps it isolated, which is why the download warns you.
## A worked example [#a-worked-example]
Your department needs to review a customer briefing, but the rest of the company does not. Set **Direct access** to your department, copy the version link, and send it in the team channel. Later, if the structure is useful as a reusable template, publish a clean version to the Marketplace. The review link and the reusable listing remain separate.
## What to check [#what-to-check]
* **A coworker says the link is dead.** The audience is still **Private**, or they are not in the department you shared with, or they are signed into a different workspace.
* **You have no department to choose.** Ask an admin to put you in one. Direct access has nothing to widen to until then.
* **Share is unavailable for a moment.** A version you just saved has to finish committing before it can be shared, so the link cannot go out under the wrong label. Wait a beat and reopen the dialog.
* **An old link says Company.** Company wide direct access grants made before this rule still open, but you cannot create new ones. Move that artifact to the Marketplace instead.
# Create your workspace (/docs/getting-started/create-your-workspace)
By the end of this page you will have a workspace, a named Talent with a working style
you chose, a channel to reach it on, and an active trial. No card required.
## Before you start [#before-you-start]
Have your company website URL handy. Setup uses it to draft your company profile so you
are confirming details instead of typing them from scratch.
## Steps [#steps]
1. Go to the sign-up page and create your account with your **work email**. The email you
use here becomes the workspace owner, and the owner is the only person who can later
change the plan or the card. Use the address you want on the bill.
2. Name your workspace. Use your company name. This is what teammates see when they are
invited, so avoid personal nicknames.
3. Confirm your company profile. Setup drafts what it can from your website, then asks
you to confirm two things: your **website** and your **industry**. Optionally open
"Add more detail" to add a company description, company size, the tools you use, and
the manual work you want off your plate. That detail is not busywork. It is the first
thing your Talent knows about you, and it makes early results noticeably better.
4. Name your Talent and give it a face. Pick something short you will actually type. Then
choose a working style: **Balanced**, **Concise**, or **Coach**. Concise gives you the
answer and stops. Coach explains its reasoning as it goes. You can change this later
in Settings.
5. Choose where your Talent reaches you. You can connect a team or mobile messaging
channel now, or skip and stay in web chat and add a channel later from
**Connections**. Most teams get more out of TalentOS once it lives in the messenger
they already have open all day.
6. You land on **Home**. The composer is the job. Ask for something real.
## What your trial gives you [#what-your-trial-gives-you]
Your trial is card-free and starts automatically. It includes **5,000 credits shared
across the entire workspace**, not per person. Credits are how AI work is measured: a
chat turn, a Routine run, a web search, a generated document. The balance is on the
**Billing** tab in **Settings**, and admins get a fuller breakdown under **Usage**.
There is no free plan. When the trial credits run out, or when you want teammates to
keep working, the workspace owner picks a plan. Paid plans start at $29 per seat per
month, or $23 on annual billing. See
[Billing](/docs/billing/plans-and-pricing).
## A good first hour [#a-good-first-hour]
* Connect the one tool your recurring work depends on most, from **Connections**. For
most teams that is email, calendar, or team chat.
* Ask your Talent for something you would otherwise do by hand right now. Not a test
question. Real work.
* Open **Settings**, then **Activity**, and read the receipt for that run. Seeing what it
actually did builds the trust you need to schedule it.
* Then do [Your first Routine](/docs/getting-started/your-first-routine).
## A worked example [#a-worked-example]
An eight person operations team creates a workspace called Northstar Ops. The owner
confirms the company website and industry, names their Talent Nova, chooses the Concise
working style, and connects a calendar. From Home they ask Nova to list tomorrow's
meetings and flag anything missing an agenda. The result is useful, so they turn that
same job into the Routine in the next guide.
## What to check [#what-to-check]
* **You were sent to a "no workspace" screen.** Someone has already invited you. Accept
that invitation instead of creating your own workspace, or you will end up alone in a
separate workspace that your colleagues cannot see. See
[Invite your team](/docs/getting-started/invite-your-team).
* **Setup could not read your website.** Fill in the industry yourself and continue. The
profile is editable afterwards and nothing is blocked by it.
* **Your Talent says it cannot reach a tool.** It is telling the truth rather than
guessing. Open **Connections** and connect that app, then ask again.
* **A workspace chat channel is already connected elsewhere.** A shared chat workspace
pairs with one TalentOS workspace. Ask an admin which workspace is correct.
* Still stuck? Email **[support@talentos.so](mailto:support@talentos.so)** with your workspace name.
# Invite your team (/docs/getting-started/invite-your-team)
Inviting people is the easy part. The part that matters is what happens in their first
ten minutes, because a teammate who gets one real result keeps using TalentOS and one who
just pokes at a chat box does not.
## Before you invite anyone [#before-you-invite-anyone]
Create at least one **department** first. Every member belongs to a department, and the
invite form will ask you to pick one. If you have no departments yet, create them from
**Team**. Departments can match your real teams: Marketing, Operations, Client Services.
## Send the invitations [#send-the-invitations]
1. Open **Team** from the Manage section of the sidebar.
2. Select **Invite to workspace**.
3. Paste the email addresses. You can separate them with commas, spaces, or new lines, so
pasting a column out of a spreadsheet works.
4. Pick a **Role**. Most people should be **member**. The **admin** option appears only
for the workspace owner.
5. Pick a **Department** for the people you are inviting. Members always hold one.
6. Send. Each person gets an email with a join link.
Only the workspace **owner** can invite admins, appoint or demote admins, or change an
admin's department. Any **admin** can invite members, resend or revoke a member
invitation, remove a member, and move members between departments.
## Which role to give people [#which-role-to-give-people]
* **Member**: the full product. Their own Talent, their own connected tools, their own
Routines, Skills, and Artifacts, and everything shared with them. This is the default
and it is the right answer for almost everyone.
* **Admin**: everything a member has, plus governance. Admins manage the team directory,
departments, connections, workspace settings, and API keys, and can view billing.
* **Manager**: not a separate invite option. It is an added seat you grant later from
**Team**, on top of someone's existing role, to give them oversight of a department.
* **Owner**: exactly one person, set when the workspace is created. Only the owner can
start, change, or cancel a plan or update the card.
## What your teammate does next [#what-your-teammate-does-next]
They click the link, sign in, and go straight into a short setup: name their Talent, pick
a face and a working style, and choose whether to connect a messaging channel.
Then they land on Home.
Send them these three things along with the invite. It is the difference between adoption
and a dead seat.
1. **Connect the one tool your work depends on**, from **Connections**. Usually email,
calendar, or team chat.
2. **Ask your Talent for something real** you would otherwise do by hand right now.
3. **Turn that into a Routine.** Point them at
[Your first Routine](/docs/getting-started/your-first-routine).
## Seats and cost [#seats-and-cost]
A seat is any **active** person in the workspace, including the owner and every admin.
Pending invitations are not seats and are never billed. The count updates when someone
accepts, and again when someone is removed. Your credit allowance is pooled across the
whole workspace rather than divided per person, so a heavy user and a light user share
one pool.
## A worked example [#a-worked-example]
An agency owner creates Client Services and Operations departments, then invites six
people in one batch. Five are members placed in their department and the operations lead
is invited as an admin. The owner sends each person the Morning brief Routine guide with
the invitation. By the end of the day, four people have accepted, so billing shows four
new active seats while the two pending invitations cost nothing.
## What to check [#what-to-check]
* **They created their own workspace by mistake.** This is the most common failure. If
they sign up with a **different** email than the one you invited, nothing connects the
two and they end up alone in a separate workspace. Ask them which address they used,
revoke the old invitation, and invite that exact address.
* **The invitation expired.** Revoke it from the invitation row in **Team** and send a
fresh one. Do not resend the old link.
* **They never got the email.** Check spam, then confirm the address has no typo. Revoke
and re-invite if it does.
* **You cannot see the Admin role option.** You are an admin, not the owner. Ask the
workspace owner to send that invitation.
* **You cannot finish the invite because no department is listed.** Create a department in
**Team** first, then invite.
# What is TalentOS? (/docs/getting-started/what-is-talentos)
TalentOS gives every person on your team an AI teammate, called a **Talent**, that does
real work with the tools you already use. Not a chat window that gives you advice, but a
teammate that reads your inbox, pulls the numbers, writes the document, and delivers it
to your team chat at 8:00 on a Monday morning without being asked.
The company gets something a personal chatbot cannot give it: a shared, governed brain.
What one person figures out becomes something the whole team can run.
## The loop [#the-loop]
1. You ask your Talent for real work, or you put that work on a schedule as a **Routine**.
2. It uses **your** connected email, calendar, documents, CRM, and team chat under
governance, with approvals for anything consequential and a receipt for every run.
3. Results land where you already are: web chat or a connected messaging channel, plus
**Artifacts** you can download as DOCX, PDF, PPTX, or XLSX.
4. Work one person perfects gets published to the company **Marketplace**, and a teammate
adopts it in one click.
## A worked example [#a-worked-example]
An agency account manager has a Friday ritual: open the CRM, check which deals moved,
skim the client spreadsheet, and write a status note in the client chat channel. It
takes 40 minutes.
In TalentOS she connects the CRM, spreadsheet, and chat tools once, asks her Talent to do it
once so she can check the output, then saves it as a Routine that runs every Friday at
4:00 pm and posts to the client channel. She reads it, edits a line, and sends it. The
40 minutes becomes 4. When a teammate takes over the account, she publishes the Routine
to the Marketplace and he adopts it.
That is the whole product. Everything below is in service of it.
## Where things live [#where-things-live]
The left sidebar is the map:
* **Home**: talk to your Talent and see what needs you.
* **Workspace**: **Routines** (repeatable work, on a schedule or whenever you ask),
**Skills** (portable capabilities your Talent can reuse), **Artifacts** (your private,
versioned documents).
* **Marketplace**: reusable work shared across your company.
* **Manage**: **Memory**, **Connections**, **Team**, **Settings**. Settings also holds
**Activity**, the durable record of every run.
The web console is for understanding and governance. Messaging channels are for attention
and quick yes or no approvals. Every run leaves exactly one receipt in Activity, and that
receipt is the canonical account of what happened.
## Who it is for [#who-it-is-for]
Teams of roughly 5 to 50 people who live in team chat and a browser, have
recurring work worth delegating, and are not going to install a desktop app or configure
a command line. Marketing, operations, agencies, professional services, and ops-heavy
small businesses.
## Roles [#roles]
There are four: **owner** (one per workspace, the only person who can change the plan or
the card), **admin** (governs the workspace, invites people, manages connections and
settings), **manager** (an added seat that gives someone oversight of a department), and
**member** (everyone else, with the full product).
## What to do next [#what-to-do-next]
If you only do one thing this week, do [Your first
Routine](/docs/getting-started/your-first-routine). One working automation is what turns
a trial into a habit.
## What to check [#what-to-check]
Before you expand the rollout, confirm one real job can run end to end: the right tools
are connected, the result is useful, the Activity receipt matches what happened, and a
Routine can repeat it. If any one of those is missing, fix that before inviting the rest
of the team.
# Your first Routine (/docs/getting-started/your-first-routine)
A **Routine** is repeatable work your Talent runs on a schedule, or whenever you ask. This
page walks one job all the way from "I do this by hand every week" to "it arrives in
team chat before I open my laptop." Give it about fifteen minutes.
## 1. Pick the right job [#1-pick-the-right-job]
Good first Routines share three traits: you already do them by hand, they have the same
shape every time, and the output is a summary or a draft rather than an irreversible
action. Skip anything that sends email to customers or changes records on the first try.
## A worked example [#a-worked-example]
We will use one job throughout: **a weekday morning brief** that reads your calendar and
inbox and posts to your direct-message channel.
## 2. Connect the tools it needs [#2-connect-the-tools-it-needs]
Open **Connections** and connect your calendar, email, and team chat. Your
Talent uses your own connected accounts, so it sees exactly what you see and nothing
more. If a tool is missing, your Talent will say so rather than invent an answer.
## 3. Run it once, by hand [#3-run-it-once-by-hand]
Go to **Home** and ask for it in the composer. Be specific about scope and format:
> Summarize my calendar for today and my unread email from the last 18 hours.
> Group it as: meetings, needs a reply today, and can wait. Keep it under 200 words.
> Do not send anything.
Read what comes back. This is the step people skip, and it is the step that decides
whether the Routine is worth having.
## 4. Read the receipt [#4-read-the-receipt]
Open **Settings**, then the **Activity** tab. Every run leaves exactly one receipt there,
and that receipt is the canonical record of what happened: the run, what it did, and what
it produced. Admins get a wider view under **Usage**, which breaks activity down by member
and by credits spent.
Check that it actually used the tools you expected. If the brief looks thin, the fix is
almost always the instructions, not the model.
## 5. Tighten the instructions [#5-tighten-the-instructions]
Rewrite the prompt with whatever went wrong fixed: name the calendar, exclude newsletters,
set the word count, say what "urgent" means to you. Run it once more. When you would be
happy to receive that output at 8:00 am, you are ready to schedule it.
## 6. Save it as a Routine [#6-save-it-as-a-routine]
1. Open **Routines** under Workspace, and select **New routine**.
2. **Name**: `Morning brief`.
3. **Category**: pick one, or type a new one. Categories group the Routines page.
4. **Instructions**: paste the prompt you just tightened.
5. **When**: choose **Once**, **Hourly**, **Daily**, or **Weekly**. For a weekday brief,
pick **Weekly** and select Monday through Friday.
6. Set the **time** and confirm the **time zone**. This is the field people get wrong.
Start and end dates are optional.
7. **Delivery**: keep results in TalentOS, or send them to a connected messaging
channel. Pick your direct-message channel.
8. Save.
## 7. Prove it works before you trust it [#7-prove-it-works-before-you-trust-it]
On the Routine's card, select **Run now**. Do not wait until tomorrow morning to find out
that a connection was missing. Then open the Routine and check the **Schedule** panel:
next run, time zone, and delivery should all read the way you expect. **Routine history**
holds the previous runs.
You can pause, resume, edit, or delete a Routine at any time from its card or its detail
page.
## What to check [#what-to-check]
* **The Routine shows "Needs attention".** Something it depends on is broken, usually a
connection that expired. Open the Routine and check the **Connections** panel, then
reconnect that app.
* **Nothing arrived at the scheduled time.** Check three things in order: the Routine is
not **paused**, the **time zone** is yours and not a default, and the **Activity** tab
shows a run at all. No receipt means it never ran. A receipt with a thin result means
it ran and the instructions need work.
* **It ran but posted nowhere.** Check the **Delivery** line. A Routine with only web
delivery leaves its result in Routine history and does not message you.
* **It stopped running after working for weeks.** Check the workspace credit balance
under **Settings**, then **Billing**. Admins can see the same thing in more detail under
**Usage**. When the pooled balance is empty, work pauses.
* **The output drifted.** Instructions are free to edit. Change them, then **Run now** to
confirm before the next scheduled run.
# Approve what your Talent does (/docs/governance/approve-what-your-talent-does)
Your Talent does not ask permission for everything, and it does not act
unsupervised either. Actions fall into three postures, and knowing which is
which stops both of the usual complaints: constant interruptions, and surprise.
Approval cards appear wherever you are working, in the web console and in your
connected messaging channel. Every action leaves a receipt in **Activity**, approved
or not.
## Never asks [#never-asks]
Safe, reversible work inside its own space. Your Talent does it and tells you
what it found.
* Reading anything you connected. Searching email, pulling a CRM pipeline,
opening a spreadsheet.
* Running code in its own sandbox to work something out.
* Keeping its own notes about you.
* Writing a private document that has no schedule on it.
* Creating, editing, running, or deleting one of your own Routines.
## Asks once for this conversation [#asks-once-for-this-conversation]
A recoverable write in one of your connected accounts. Updating a knowledge
page, commenting on an issue, or adding a row to a database.
The card offers three answers: **Approve**, **Decline**, and **Allow for this
conversation**. The third one covers the same action on the same account for the
rest of that thread, so a long piece of work does not raise the same card twenty
times.
A different action asks again. A different account asks again. A new
conversation asks again. Grants die with the conversation, by design.
This only applies in a web chat conversation. A messaging-channel turn, and a
step inside a Routine, ask every time, because there is no conversation that
can end and close the grant.
## Asks every single time by default [#asks-every-single-time-by-default]
Six kinds of work always ask, however many times you have said yes before.
1. **Anything sent to a person.** An email reply or a team-chat message.
2. **Anything that books calendar time.** Creating or moving a calendar event
spends other people's hours, and deleting the event gives
none of them back.
3. **Anything that moves money.** Invoices, payments, ad spend.
4. **Anything that deletes.** Any delete, archive, clear, or purge, on any
connector.
5. **Anything that changes who can reach your data.** Sharing a drive file,
adding a project member, or setting a secret.
6. **Anything that arms a standing commitment.** Putting a Routine on a
schedule, so it keeps firing after you have walked away.
Custom connectors you registered yourself always ask too. So does proposing a
fact into department memory, because the whole department reads it.
## A worked example [#a-worked-example]
You ask for a weekly client update: pull the CRM pipeline, draft the summary,
email it to the client.
Reading the CRM: no card. Drafting the summary: no card. The email send: a card,
showing exactly what will go out, to whom. Approve it and it goes. That is one
decision on a job with a dozen steps.
## Change the defaults [#change-the-defaults]
Open **Settings** and find the action policies. Each governed action can be set
to:
* **Ask**, the default. You see exactly what will happen before it happens.
* **Auto**, run it without a card. It still leaves a receipt.
* **Off**, block it. Your Talent will tell you the setting stopped it.
If you run that same weekly update every Monday and you trust the draft, set the
email send to **Auto** and the Routine completes without you. Set anything you
never want it touching to **Off**.
## What to check [#what-to-check]
**"Your admin requires approval for this action."** Your workspace is set so
that action always asks. You can leave it on Ask, but you cannot set it to Auto.
**A scheduled Routine finished without doing the last step.** It parked on an
approval and is waiting for you. Approve it from the bell, or set that eligible
action to Auto if it runs every week.
**Your Talent says it was stopped by a setting.** That action is set to Off.
Change it in Settings.
**It keeps asking for something you approve constantly.** Approving repeatedly
never turns into permission. Only the switch in Settings does.
# What admins can and cannot see (/docs/governance/what-admins-can-and-cannot-see)
This is the question every team asks before rolling TalentOS out, and it
deserves a precise answer rather than a reassuring one.
**Visibility is artifact states, never behavior.** An admin can see what exists
in the workspace and what it cost. An admin cannot see how any individual
worked, what they asked, or what their Talent said back. That line is enforced
on the server, not in the interface, and it applies to the owner and to managers
as well.
Tell your team this before they connect their mail.
## What an admin can see [#what-an-admin-can-see]
* **Published work.** Anything anyone published to the **Marketplace**, plus
department and company memory. It was shared on purpose.
* **The workspace roster.** People, roles, departments, pending invitations, and
each person's Talent by name with its status.
* **The plan and the pool.** Seats, plan, renewal, and the shared credit balance.
* **What work cost.** The usage dashboard in **Settings** shows workspace spend
broken down by member and by routine.
## What no admin can see [#what-no-admin-can-see]
* **Chats.** Nobody else reads your conversations with your Talent, in the web
console or a connected messaging channel.
* **Personal memory.** Your Talent's memory of you is yours. Admins are not an
exception to that.
* **Your connected accounts.** Connecting email connects it to your Talent,
not to the workspace. An admin never reads your mail through TalentOS.
* **What your Talent produced.** Private Artifacts and private Routines and
Skills stay private until you publish them.
* **Anything a run said.** No summary, no error text, no instructions, no
context. The usage dashboard is billing information and nothing else.
Spend is the one thing that crosses the line, because someone has to pay the
bill. The owner and any admin can see how many credits each member used. They
cannot see what those credits were spent doing.
## The usage dashboard, precisely [#the-usage-dashboard-precisely]
Open **Settings** and go to usage. Owner and admins see it. Members do not.
For each member it shows how much they used, when, whether runs succeeded, and
which model lane ran. For each routine it shows what that routine cost.
Routine names are where this gets subtle, and the product is careful about it. A
routine only shows its real name if it was published to the whole workspace.
Anyone else's private routine appears as **Private routine**, with its cost
attached. So an admin can see that one member is burning a lot of credits on a
private routine and can go and ask them about it, which is the honest amount of
information to hand someone who signs the invoice.
Removed members stay in the list, labelled. Their spend was real spend, and
dropping them would stop the member column adding up to the workspace total.
## Receipts, and who they belong to [#receipts-and-who-they-belong-to]
Every run leaves one receipt in **Activity**, inside Settings. Receipts are the
canonical record of what happened, and your Activity is your own history, not a
feed an admin browses. The bell in the header is a bounded inbox of things
needing you right now, such as an approval or a low credit warning. It is not a
history and not a management view of anyone else.
## A worked example [#a-worked-example]
A marketing admin sees that one member used 1,200 credits on a private Routine
this week. The dashboard labels the work **Private routine**, so the admin can
ask about the cost but cannot open the instructions, results, or conversation.
If the member publishes the Routine to the Marketplace, its shared name and
version become visible to the workspace.
## What to check [#what-to-check]
**A member says they can see spend and you did not expect it.** They are an
admin. Check their role on the **Directory** tab.
**An admin wants a per person report of what was actually done.** It does not
exist and will not be added by a setting. Ask people to publish the work you
want visibility over to the Marketplace, which is the sanctioned path.
**A routine you own shows as "Private routine" to your admin.** That is correct
and expected. Publish it to the workspace if you want it named.
**Usage numbers look short.** The dashboard reads live over a bounded recent
window, and it tells you when a view was truncated rather than quietly showing
you less. Narrow the range and read it again.
# What only the owner can do (/docs/governance/what-only-the-owner-can-do)
Every workspace has exactly one owner: the person who created it. The owner is
an admin too, with one extra layer on top. That layer is deliberately small, and
it is only two things: money, and who else gets to govern.
Knowing this list saves the usual half hour of an admin clicking around
wondering why a button is not there.
## Owner only [#owner-only]
**Money.**
* Start a plan or go through checkout.
* Change plan, upgrade, downgrade, cancel.
* Update the card, through the secure billing portal.
**Admin appointment.**
* Promote a member to admin, or change an admin back to a member.
* Invite an admin.
* Revoke or resend an **admin** invitation.
* Remove an admin.
* Change an admin's department.
**Workspace identity.** Renaming the workspace and finishing the company setup.
## What any admin can do [#what-any-admin-can-do]
The split is narrower than people expect, so be precise about it.
* **Buy credit top ups.** Any admin can top the workspace up when it runs low.
That is not gated to the owner, so nobody has to wait for one person to unblock
the team mid week.
* **View billing.** Any admin sees the plan, the status, the renewal, and the
credit pool.
* **Run the workspace.** Invite and remove members, departments, manager seats,
workspace settings, and workspace spend.
Members see billing too, read only. Owner managed billing is simply the normal
state for a member, so they get the plan and the balance and no controls.
A non-owner admin does not get a billing button that fails. They get a note
saying billing is owner managed, in place of the controls. A rendered control
you cannot use is worse than no control.
## The owner seat is protected [#the-owner-seat-is-protected]
**The owner cannot be removed.** Any attempt returns "The workspace owner can't
be removed." The owner also cannot be demoted, including by themselves, so
nobody can accidentally leave a workspace with no owner.
If the owner is leaving the company, contact support before their last day, not
after their account is gone.
## When the owner is not around [#when-the-owner-is-not-around]
* **Out of credit mid sprint.** Any admin buys a top up. No owner needed.
* **Card expiring.** Only the owner can update it. Give them notice, because a
lapsed payment puts the workspace into read only, and read only blocks
invitations, removals, role changes, and other writes for everyone.
* **You need another admin urgently.** Only the owner can appoint one. An admin
can invite the person as a member now, and the owner promotes them later.
## A worked example [#a-worked-example]
A 25 person agency runs low on credits while its founder is travelling. The
operations admin buys a top-up immediately, because any admin can do that. A
week later the card needs updating, so the founder opens the secure billing
portal as the workspace owner. The admin can see the billing state throughout,
but only the owner can change the card.
## What to check [#what-to-check]
**"Only the workspace owner can appoint or remove admins."** You are an admin,
not the owner. Ask them.
**"Only the workspace owner can invite an admin."** Same cause. Invite the
person as a member, then have the owner promote.
**"Promote another admin before demoting the only workspace admin."** A workspace
always keeps at least one admin. Promote someone first.
**You think you are the owner but the controls are missing.** There is one owner
per workspace, and it is the person who created it. Check whether you are in the
workspace you think you are in, using the workspace switcher.
# Who can do what (/docs/governance/who-can-do-what)
Work out which of the four seats each person needs, and you are done. TalentOS
has no fine grained permission editor to configure, on purpose.
The rule underneath all of it: **a seat gates authority, never participation.**
An admin is not a supervisor account with a different product. Everyone has
their own Talent, their own connected tools, their own memory, and their own
work. Higher seats simply add governance on top.
## The four seats [#the-four-seats]
**Member.** The everyday seat and the whole product. A member has their own
Talent, connects their own apps, writes Routines and Skills, keeps their own
memory, publishes to the Marketplace, and adopts from it. Every member sits in a
department.
**Admin.** A member, plus governance of the workspace. Admins invite and remove
members, create and archive departments, move people between them, assign
manager seats, hold workspace settings, and read billing and workspace spend.
**Manager.** Not a role. A department scoped grant that sits on top of whatever
someone already is, so a member or an admin can hold one. It adds a **Manager**
group in the sidebar for the department they run, where they curate its shared
memory. One manager per department.
**Owner.** Exactly one per workspace, and always an admin as well. The owner
holds money and admin appointment. See
[What only the owner can do](/docs/governance/what-only-the-owner-can-do).
## Pick the right one [#pick-the-right-one]
| You want them to | Give them |
| ---------------------------------------------- | --------------------------------- |
| Do the work, with their own Talent and tools | Member |
| Look after one department's shared knowledge | Member plus a manager seat |
| Invite people and run departments and settings | Admin |
| Change the plan or the card | They already need to be the owner |
Give admin to the people who actually administer, usually one or two in a team
of 20. It buys them nothing extra as a practitioner, and it does not let them
read anyone's work.
## What a Talent may do [#what-a-talent-may-do]
A Talent holds **capabilities, never a role**. It is a bounded subset of the
authority of the person who owns it, and it can never exceed them.
* It acts as its owner, with their connected accounts. Priya's Talent reads
Priya's email because Priya connected it, and it has no path to yours.
* It cannot publish work to the department or the company on its own. Widening
an audience is a decision a person makes in the console.
* Consequential actions carry an approval step. See
[Approve what your Talent does](/docs/governance/approve-what-your-talent-does).
* Talking to your Talent from a connected messaging channel gives it exactly the
same authority it has in the web console, and no more.
## A worked example [#a-worked-example]
Priya runs Client Services but does not administer the workspace. Give her a
member role plus the Client Services manager seat. She can use her own Talent,
connect her own tools, and curate that department's shared memory. She cannot
invite a new teammate or change billing, so an admin handles the invitation and
the owner handles the plan.
## What to check [#what-to-check]
**Someone is missing a section of the sidebar.** Sections are absent rather than
disabled when a seat does not include them. Check their row in **Team**,
**Directory**, and confirm the seat, then ask them to reload.
**An admin cannot change a plan.** Correct. Billing writes are owner only.
**A control is visible but refuses.** Every rule here is enforced on the server,
not just in the interface, so the refusal is the real answer. The message names
which seat the action needs.
# Add someone else's work to your workspace (/docs/marketplace/find-and-add)
The Marketplace is where you stop rebuilding things your colleagues already got right. Someone in your company perfected a weekly reporting Routine; you add it, it becomes yours, you change the parts that are wrong for you. That is the whole loop.
Open **Marketplace** in the left navigation.
## Browse [#browse]
The page has two sources, shown as tabs at the top:
* Your company's tab, named after your workspace. Routines, Skills and published Artifact versions that people you work with shared.
* **Public**. A catalog TalentOS curates, so a brand new workspace has something to start from.
Everything below the tabs works the same either way:
* Filter by type: **Routines**, **Skills** or **Artifacts**.
* Search by name.
* Sort. On your company's tab that is **Recently shared**, **Popular recently**, **Highest rated** or **Most added**.
* Narrow to **In your library** or **Not added yet** when you have lost track of what you already took.
Open any card to read it before you commit. A listing from your company also shows its rating and any comments coworkers left, which is usually the fastest way to tell a polished Routine from a first draft.
## Add it [#add-it]
Choose **Add** on the card or in the detail view.
That creates **your own private copy**. It is not a link, not a subscription, and not shared back. The original author cannot see or change your copy, and your edits never touch theirs.
## A worked example [#a-worked-example]
Someone in Ops publishes a Routine called *Weekly pipeline digest* that reads HubSpot, writes a summary, and posts it to Slack.
1. Find it on your company's tab and open it.
2. Read the steps. It uses HubSpot and Slack.
3. Choose **Add**.
4. Open **Routines**. Your copy is there, with **no schedule yet**, because a schedule is yours to decide.
5. Set it to run Monday mornings, or just ask your Talent to run it now.
6. Change the Slack channel to yours. It is your copy; nothing breaks upstream.
An added Routine or Skill runs on **your** connected tools, under your permissions. It cannot reach anything you have not connected, and adopting somebody's work never hands you their access.
## Rate it and say something [#rate-it-and-say-something]
If a listing from your company saved you an afternoon, rate it and leave a comment. That is what feeds **Highest rated** and **Most added**, and it is how the next person knows which of five similar Routines to take.
## What to check [#what-to-check]
* **Add is greyed out.** You already have a copy, or the listing was removed by its author. Check **In your library**.
* **Your copy fails on the first run.** It needs a tool you have not connected. Open **Connections** and connect that app, then run it again. A Routine that reads Gmail needs your Gmail, not the author's.
* **Your company's tab looks empty.** Nobody has published yet. Anything one person builds can be shared, so this fills up as your team works.
* **A card says an artifact was deleted.** Its owner deleted the source. Nothing to add; ask them to republish.
* **You cannot find something a coworker told you about.** Department shared work is not in the Marketplace. Look under **Shared with you** on the relevant Workspace page instead.
# Share a Routine or Skill with your team (/docs/marketplace/publish-your-work)
When a Routine or Skill starts doing real work, other people probably need it too. Sharing hands them a copy they can add and run themselves.
Where it lands depends on how wide you go. Share with **your department** and it shows up for them under **Shared with you** on the Routines or Skills page. Share with **the whole company** and it becomes a Marketplace listing anyone can find, rate and add.
## Who has to approve [#who-has-to-approve]
| You want it seen by | Who approves |
| ------------------- | --------------------------------------------------- |
| Just you | Nobody. Build and run whatever you like, privately. |
| Your department | That department's manager, or any admin |
| The whole company | An admin |
If you are an admin, your own share goes straight out. If you are a member, it goes to review first, and the button says so.
## Share a Routine [#share-a-routine]
1. Open **Routines** and pick the one that works.
2. Choose **Share**.
3. Pick **your department** or **the whole company**.
4. Choose **Share** if you are an admin, or **Submit for review** if you are a member.
What actually gets shared is a **Skill copy** of the Routine's instructions, not the Routine itself. Your schedule, your connected accounts and your private setup stay yours. Whoever adds it gets the recipe and runs it on their own tools.
Improved the Routine since? Open **Share** again. It refreshes the same copy in place rather than littering the Marketplace with near duplicates.
## Share a Skill [#share-a-skill]
Same shape, from the **Skills** page: open the Skill, choose **Share**, and answer **Who can see this skill?** by picking a department or the company. Members submit; admins share.
## Share an artifact [#share-an-artifact]
An artifact publishes from its own **Share** dialog, in the **Marketplace** section. That lists one frozen version. See [Share an artifact with the right people](/docs/artifacts/share-an-artifact).
Nothing you share leaves your workspace. The Marketplace is your company's, and the audience you pick is the widest it can go. There is no public posting from here.
## A worked example [#a-worked-example]
You built a Routine that reads the shared Gmail inbox every morning, sorts the new messages into themes, and posts a digest to Slack. It has been solid for three weeks.
1. Open the Routine, choose **Share**, pick **the whole company**.
2. You are a member, so the button reads **Submit for review**. Choose it.
3. An admin opens it, reads the exact version you submitted, and approves.
4. It appears on your company's Marketplace tab. A colleague chooses **Add**, connects their own Gmail and Slack, and sets their own schedule.
## What to check [#what-to-check]
* **The department option is disabled.** You are not in a department. Ask an admin to assign you to one, or share with the whole company instead.
* **You submitted and nothing happened.** Reviews are done by a person. Ping your manager or an admin; the item is sitting on their Skills page marked **Needs review**.
* **It says your copy is already shared.** You have not changed the Routine since the last share, so there is nothing new to submit. Edit it first.
* **You cannot see it in the Marketplace yet.** Only approved, company wide versions are listed. Private work never appears there, and department shared work lives under **Shared with you** rather than in the Marketplace.
* **You want it taken down.** Set its audience back to private, or ask an admin. Copies people already added stay theirs.
# Approve what your team wants to share (/docs/marketplace/review-what-your-team-submits)
Members can build and privately run whatever they want. What they cannot do is widen it. The moment a member wants their work seen by their department or by the company, it comes to you.
* **A manager** approves sharing into their own department.
* **An admin** approves sharing into a department or across the whole company.
## Find what is waiting [#find-what-is-waiting]
1. Open **Skills** in the left navigation.
2. Set the filter to **Needs review**. Pending items also sort to the top on their own, and each carries a **Needs review** badge.
3. Open one.
Routines arrive here too. When somebody shares a Routine, the product mints a Skill copy of its instructions and that copy is what you review, so everything pending is in one place.
## Read it before you approve [#read-it-before-you-approve]
You are approving one exact version, not a living document. Read the instructions and the success criteria the author wrote, and ask the practical questions:
* Does it name the right tools, channels and audiences for the people who will run it?
* Does it say clearly enough what "done" looks like that somebody else can rely on the result?
* Does anything in the text contain information that should not travel further, such as a customer name or an internal figure?
Nothing about approving grants access to the author's private setup. Whoever adds this later runs it on their own connected accounts, under their own permissions.
## Approve or send it back [#approve-or-send-it-back]
* Choose **Approve and share**, then confirm the audience that was requested. That version becomes visible to the department or the company and, at company scope, becomes a Marketplace listing.
* To decline, write a note in the review feedback box and choose **Send feedback**. The author sees your note on the Skill and can fix it and resubmit. Feedback is required, so nobody gets a silent rejection.
If the author edits the Skill while it is waiting, the pending request is dropped, because the version you were asked to approve no longer exists. They need to submit again. This is the point: you can never approve one version and publish another.
## A worked example [#a-worked-example]
A member on the Marketing team submits *Weekly competitor scan* to the whole company. You open it, see it names a specific Notion database that only Marketing can reach, and decide it is not company material as written.
Write: *"Great Routine. Point it at a page the whole company can open, or resubmit this to Marketing only."* Choose **Send feedback**. They resubmit to their department, their manager approves, and it is live for the people it actually suits.
## What to check [#what-to-check]
* **Approve and share is not there.** The item is not pending, or it is not yours to govern. A manager cannot approve a request aimed at another department or at the whole company.
* **You cannot find a submission you were told about.** Check that you are on **Skills**, not Routines, and that the **Needs review** filter is on.
* **Send feedback is disabled.** The note is empty. Type something first.
* **The author says it vanished.** They edited it after submitting. Ask them to submit the new version.
# Get a fact into company memory (/docs/memory/add-to-company-memory)
Company memory is the workspace's shared brain. Everyone's Talent reads it, so nothing lands there by accident. A human always puts it there or approves it.
Open **Memory**, then the tab with your workspace name.
## If you are an admin or the owner [#if-you-are-an-admin-or-the-owner]
You curate directly. What you write is stored as approved knowledge, because deciding is your job.
1. Click **Add to memory**, then **Write a memory**.
2. Choose the type: **fact**, **decision** or **SOP**.
3. Write one declarative sentence, up to 200 characters.
4. Save.
To change an existing page, open its section, find the row and click **Edit page**. To take one out of circulation, archive it. Archive is reversible and no accepted page is ever deleted in place.
## A worked example [#a-worked-example]
Finance keeps getting surprised by refunds. An admin adds a decision: "Refunds above $500 need written finance sign off before they are issued." From the next turn on, a member asking their Talent to draft a refund email to a customer gets that condition applied without anyone quoting the policy at it.
## If you are a member [#if-you-are-a-member]
You propose. Same button, same form. Your submission becomes a proposal for an admin to review, and the button says **Propose** rather than **Add**.
You can propose four things:
* **Add a new page** for knowledge that is missing.
* **Change current page** when a page is out of date.
* **Archive current page** when a rule no longer applies.
* **Restore archived page** when one was retired too early.
You will not see the review queue or who else proposed what. You do get told when your proposal is approved or rejected.
## Reviewing proposals [#reviewing-proposals]
Admins and the owner see a **Review** section on the company memory page.
1. Open the proposal. It shows the proposed action, the page it targets, and the page's current text side by side.
2. Correct the wording if it needs it, and set the type.
3. Click **Approve revision**, or **Reject**.
The version you approve is the version that gets stored, wording and type included. Rejecting keeps the original proposed text on record so nothing is quietly rewritten.
Two badges matter. **Stale target** means the page moved on after this was proposed, so approving would clobber a newer version. Re-check the current text before deciding. **Safety blocked** means the content tripped the credential filter and cannot be approved as written.
Nothing your Talent produces becomes approved company knowledge on its own. Work derived candidates arrive in the same review queue as member proposals and wait for a human.
## The sections you are writing into [#the-sections-you-are-writing-into]
Company overview, Customers and product, Strategy and priorities, Organization and decision rights, How we operate, Policies systems and risk, and History and important decisions once it has content.
Write for the section. A pricing rule goes in Strategy and priorities, not in Company overview where nobody looking for pricing will open it.
A shared space holds 100 current pages. The **Knowledge capacity** strip on the page shows how full it is. When it is full, new commits are reported as skipped rather than silently dropped, so archive retired pages instead of letting them sit.
## What to check [#what-to-check]
* **You see Propose where you expected Add.** You are a member in this workspace. Ask an admin to review your proposal, or to make you an admin.
* **Approval failed with a conflict.** Someone else changed the page first. Reopen the proposal against the current version and approve again.
* **The save was refused for content.** The credential filter rejects anything that reads like a key, token or card number, including prose that merely discusses them. Rewrite without naming the credential.
* **Nobody is reviewing.** Pending proposals sit in the bell for whoever is authorised to decide them. If that person left the workspace, another admin can clear the queue.
# Curate your department's memory (/docs/memory/curate-department-memory)
Department memory is for the things true of one team and not of the whole workspace: what it owns, how it runs its week, which tools it standardises on.
Open **Memory** and pick the tab named after your department. Where that tab sends you depends on what you hold.
## Who can do what [#who-can-do-what]
| You are | What you get |
| ---------------------------- | --------------------------------------------------------- |
| A member of the department | Read everything approved, and propose changes |
| A manager of that department | Direct editing, plus the review queue for that department |
| An admin or the owner | Direct editing and review for any active department |
A manager seat is a grant on top of your normal role. A plain member who manages their own department gets the curator view, not the propose only one.
## Curate a department page [#curate-a-department-page]
1. Open the department tab.
2. Click **Add to memory**, then **Write a memory**.
3. Pick the type: **fact**, **decision** or **SOP**.
4. Save. Your text is stored as approved knowledge for that department.
To change something, open the section and click **Edit page**. To retire a page, archive it. Every accepted change writes a new version and the old one stays in history.
## A worked example [#a-worked-example]
A five person revenue team keeps losing track of inbound leads. Their manager adds an SOP: "Every inbound lead is logged in HubSpot within one working day, owner assigned before end of day." A rep who then asks their Talent to draft a follow up gets the ownership step included, and a Routine that summarises the week can be written against the same rule rather than someone's memory of it.
## Propose as a member [#propose-as-a-member]
Members use the same **Add to memory** button, and the confirm step says **Propose**. You can propose an add, a change, an archive or a restore. You never see the pending queue or who proposed what. You are told when your proposal is decided.
## Review proposals [#review-proposals]
Managers of the department, admins and the owner see a **Review** section on the department page.
1. Open a proposal. It names the action, the page it targets, the current text, and whether the target is still current.
2. Fix the wording and the type if needed.
3. **Approve revision** or **Reject**.
A **Stale target** badge means the page changed after the proposal was written. Read the current version before approving. A **Safety blocked** badge means the text tripped the credential filter and cannot be approved as written.
## The sections you are writing into [#the-sections-you-are-writing-into]
Overview and mandate, Priorities, People and ownership, How we work, Tools and standards, and Challenges and dependencies once it has content.
Company memory wins on conflict. A department page can add detail to a company rule, sharpen it or say how your team applies it. It cannot override it. If a company page is wrong, fix it there instead of writing a contradicting department page.
Members read the memory of their **current active department** only, and that scope updates the moment someone moves them. There is no way to read another department's memory as a plain member, and an archived department stops serving pages entirely.
## What to check [#what-to-check]
* **No department tab.** You are not assigned to an active department. An admin fixes that from **Team**.
* **You see Propose but you manage this team.** Your manager seat is on a different department, or it was not assigned. Ask an admin to check your seat under **Team**.
* **A teammate cannot see a page you just wrote.** Confirm they are in the same department, and that you saved to the department tab rather than to **You**.
* **A page vanished after a reorganisation.** The department was archived, or you were moved. Ask an admin which department now owns that knowledge, and move it to the company space if it applies to everyone.
# Turn a document into memory (/docs/memory/import-a-document)
You have a brand guide, an onboarding checklist or a pricing sheet, and you want your Talent to work by it. Import it once. Every memory space accepts documents.
## Import it [#import-it]
1. Open **Memory** and pick the space you are filling: **You**, your department, or your workspace.
2. Click **Add to memory**, then **Import a document**.
3. Drop in one file. PDF, DOCX, XLSX, PPTX, Markdown or plain text, up to 10 MB. PDFs are read up to 50 pages.
4. Wait while the file is read and classified.
5. Review the candidate lines. Edit any of them, retype them, or untick the ones you do not want.
6. Commit. The button says **Save** with a count if you can write directly, or **Submit for review** with a count if you are a member proposing.
## A worked example [#a-worked-example]
An agency imports its 9 page client onboarding SOP into department memory. It comes back as about a dozen lines: kickoff happens within three working days, the Notion project page is created before the kickoff call, invoices go out on the first of the month. The manager deletes two lines that are already stale, fixes the wording on one, and saves. From then on a Talent drafting a kickoff email or a Routine building the weekly client update works from the same rules.
This is deliberately lossy. It captures the durable, reusable knowledge in the document, not the document verbatim. Each line is capped at 200 characters. Read every candidate before you commit: they are model generated, so a line that is subtly wrong is your responsibility to catch, not the reviewer's downstream.
## Keeping the source file [#keeping-the-source-file]
On your personal space you can also tick **Keep document**, which stores the extracted text as a private source document you can point your Talent at later. It is not injected into every turn. Your Talent reads it when you ask for it by name.
Shared spaces do not keep the file. They record where the facts came from, so the source shows up as provenance on each row.
## What the result line means [#what-the-result-line-means]
After committing you get a one line summary. It can say any of:
* **Saved** for lines written straight into the space.
* **Submitted for review** for lines that became proposals because you are a member.
* **Already known** for lines that duplicate something the space already holds.
* **Skipped, memory full** for lines that hit the capacity ceiling. Those are reported, never dropped in silence.
* **Skipped** for lines the safety filter refused.
A batch never fails halfway. You always get a per line outcome.
Capacity is real. A shared space holds 100 current pages. Your personal memory has one 50 active page owner ceiling across every personal memory domain, not one allowance per section or surface. If a large import comes back mostly as "skipped, memory full", archive retired pages first, then import again.
## What to check [#what-to-check]
* **"Could not read that file."** It is password protected, or it is a scan with no text layer. Export a text based version and try again.
* **The file is rejected on size.** The ceiling is 10 MB for one file. Split it, or import the section that actually holds the rules.
* **You get two candidates instead of twenty.** The document is mostly narrative, tables or images. Documents written as rules distill well. Slide decks and prose reports often do not.
* **Nothing new to save.** Everything in the document is already in the space. That is a pass, not a failure.
* **You imported into the wrong space.** Archive the rows in the space you filled, then import again from the correct tab. Nothing moves between spaces automatically.
# Put the right facts in Memory (/docs/memory)
Your Talent does better work when it stops asking you the same question every week. Memory is where you put the answer once.
Open **Memory** in the sidebar. You land on a tab row with three spaces: **You**, your department, and your workspace name.
## The three spaces [#the-three-spaces]
| Space | Who can read it | Who can change it |
| ------------------ | --------------------------- | ---------------------------------------------------------------- |
| **You** (personal) | Only you | You |
| **Department** | Everyone in that department | Managers of that department, admins, the owner. Members propose. |
| **Company** | Everyone in the workspace | Admins and the owner. Members propose. |
Personal memory is private to you. Not your manager, not an admin, not the workspace owner. There is no screen anywhere in TalentOS that shows one person's personal memory to another. Do not use it for anything a colleague needs to see: that belongs in department or company memory.
## What to put where [#what-to-put-where]
* **You**: how you like work delivered, the clients you own, the spreadsheet you always report from, the people you work with most.
* **Department**: what your team is responsible for this quarter, the tools it standardises on, who owns which account, the way it runs its weekly cycle.
* **Company**: policies, pricing rules, positioning, approval thresholds, decisions that everyone must follow.
Company knowledge is workspace wide and wins when it conflicts with a department page. A department can add detail to a company rule. It cannot overrule it.
## How a space reads [#how-a-space-reads]
Each space is a document, not a list of rows. You see a short paragraph per section first, for example **How we operate** or **Priorities**. Open a section and the paragraph is replaced by the individual facts behind it, each with its date and where it came from.
Both the paragraph and the facts are real. The facts are the record. The paragraph is the readable summary your Talent is handed.
On a work turn your Talent receives at most **two** relevant sections of shared memory, each with at most **two** supporting excerpts. Memory informs a turn, it is not loaded into every turn wholesale. That is why a short, sharply written section beats a long one: a 40 line policy page competes with itself.
## A worked example [#a-worked-example]
You prefer concise weekly updates, your department owns the launch calendar, and the whole company requires approval before quoting a customer. Put the delivery preference in **You**, the launch ownership rule in **Department**, and the customer rule in **Company**. Your Talent can then apply the right fact without treating a personal preference as company policy.
## Start here [#start-here]
## What to check [#what-to-check]
* **The department tab is missing.** You are not assigned to an active department. An admin can assign you from **Team**.
* **You changed department and the old pages are gone.** Correct. Department memory follows your current active department and re-scopes the moment you move.
* **Your Talent ignored something you wrote.** Confirm you put it in a space your Talent actually reads for you, and that the wording names the thing it should match. A fact filed under a section nobody's work touches will rarely be selected.
# Check where a memory came from, and undo it (/docs/memory/provenance-and-undo)
Somebody's Talent said something odd, or a page contradicts what you know. Every memory in TalentOS can be traced and reversed. Here is how.
## Find the fact behind the paragraph [#find-the-fact-behind-the-paragraph]
Section paragraphs are summaries. The facts are the record.
1. Open **Memory** and pick the space.
2. Click the section heading to open it. The paragraph is replaced by the individual rows that support it.
3. Each row shows its text, its date, and where it came from.
Personal rows say **Added by you**, **Captured from your Talent**, or **Learned from your conversation**. Shared rows carry the proposer, the decision, and the source document or integration when one was recorded.
You will never see raw conversation text, prompts, run identifiers or another member's personal memory as provenance on a shared page. Personal context is never company provenance.
## Read the history [#read-the-history]
Open a row and you get its versions in order. Every accepted change writes a new version rather than editing the old one in place, so history is complete by construction. Only the current version can be acted on.
Links of the form `?memory=...` stay valid across edits. An old link resolves to the safe current version of that page and tells you the version you asked for was superseded.
## Undo a shared page [#undo-a-shared-page]
* **Archive** it if it is out of date. Retrieval stops immediately, and the page moves to **Archived**.
* **Restore** it if you retired it too soon. Restoring writes a new current version.
Archiving is reversible and nothing accepted is ever hard deleted through these screens. Managers curate their own departments, admins and the owner curate any active department and the company space, and members propose archive and restore the same way they propose a change.
## Undo a personal memory [#undo-a-personal-memory]
You have three options on your own rows:
* **Edit** to correct the wording. The old text stays in history.
* **Archive** to take it out of use. Reversible.
* **Forget** to remove it permanently.
**Forget is destructive and it is meant to be.** It deletes the fact and its entire version chain, so nothing you asked to be forgotten survives in history. It also leaves a marker so the same content cannot be silently re-saved later by a background capture. Archive is the reversible one. Use Forget when the content itself should not exist.
## Export [#export]
Every space offers an export, so you can take a copy of what is stored before a clean up or an audit.
Editing supersedes, it does not remove. If a stale fact keeps showing up in your Talent's work, check that you archived or forgot the old row rather than only adding a corrected one beside it. Two contradictory rows both stay eligible.
## A worked example [#a-worked-example]
A department page still says campaign reports go out on Thursday, but the team moved them to Friday. Open the section, trace the Thursday rule to its source, and edit or archive it before adding the Friday rule. Then check history to confirm the stale version is no longer current.
## What to check [#what-to-check]
* **A row has no source label.** It was added by hand rather than captured, or its source could not be safely resolved. That is expected, not a bug.
* **Restore was refused.** The space is at its page ceiling, or a newer version already occupies that page. Archive something retired, then restore.
* **You archived a page and your Talent still used it.** Confirm you archived in the space the Talent actually reads for that person. A company page and a department page can say similar things, and company wins.
* **A member cannot archive a shared page.** Members propose archive; a manager, admin or the owner decides it.
* **You need a personal memory gone for good.** Archive is not enough. Use **Forget**.
# Teach your Talent about your work (/docs/memory/teach-your-talent)
Stop re-explaining your job at the start of every conversation. Personal memory is the set of durable facts your Talent carries into your work, and only you can see it.
## Write a memory [#write-a-memory]
1. Open **Memory** in the sidebar. You start on the **You** tab.
2. Click **Add to memory**, then **Write a memory**.
3. Type one declarative sentence, up to 200 characters. Save.
One fact per memory. "The weekly Novabrief report is built from the Google Sheets tracker, never the HubSpot export" is a memory. A three paragraph brief is not.
## A worked example [#a-worked-example]
Add one fact: `Friday campaign reports use the approved tracker and need a three-bullet executive summary.` The next time you ask for that report, your Talent can apply the source and format without asking again. If the rule becomes team-wide, propose it to department memory instead of keeping it personal.
Facts group themselves into sections such as **How you work**, **People**, **Projects**, **Documents** and **Decisions**. Anything that does not classify cleanly stays visible under **Other** rather than being filed somewhere wrong.
## Let your Talent capture it [#let-your-talent-capture-it]
You can also just say it in a private conversation: web chat, a Slack DM, Telegram or iMessage.
> Remember that Novabrief's weekly report goes out Friday morning, and that Priya reviews it before it ships.
When your Talent saves something, a short note appears beside its reply and the new fact shows up under **Recent** on your Memory page.
Your Talent never captures memory from a Slack channel, a private group, a group DM or a Routine run. In a channel you spoke to an audience, not to your Talent. Capturing nothing from a conversation is the normal result, not a failure. If a fact matters, write it on the Memory page instead of hoping it was picked up.
## Pin what must always apply [#pin-what-must-always-apply]
Open a section, hover a fact, and pin it. Pinned facts are considered on every turn. Unpinned facts are pulled in when the conversation actually relates to them.
Pin standing rules: your timezone, your reporting format, the one client whose rules differ. Do not pin everything. Pins share the same budget as everything else your Talent recalls, so a wall of pins crowds out the fact the current turn needed.
## Facts about a colleague [#facts-about-a-colleague]
Ask your Talent to remember something about a named teammate and it files that fact under that person's identity, so it still surfaces after they change their display name. A name that matches nobody in the workspace still saves, unkeyed, which is what you want for a client or a contractor. A name that matches two people is refused and your Talent asks which one you meant.
That note is still yours alone. The person it is about cannot read it, and your Talent never repeats it as if it were their own words.
## Recent is not memory [#recent-is-not-memory]
**Recent** shows short lived notes about what you have been doing lately. They fade on their own. They are not durable facts and are not a change log. If something in Recent should stick, write it as a memory.
You can hold **50 active personal memory pages**, counted for you across every personal memory domain, not per section or surface. At the ceiling nothing new saves. The bell tells you: "Memory is full. Your Talent stopped saving new memory at 50 of 50 pages. Archive or forget one to make room." That notice clears itself the moment you free a slot, so you cannot dismiss your way past a full pool.
## What to check [#what-to-check]
* **A memory would not save.** You are at 50. Archive a fact you no longer need, or use **Forget** to remove it permanently, then save again.
* **Nothing was captured from your chat.** Check it was a one to one conversation, not a channel or group. Then write the fact yourself.
* **A saved fact is wrong.** Open the section, click the pencil on the row and correct it. Editing creates a new version and keeps the old one in history.
* **Your Talent keeps using an old fact.** Editing supersedes, it does not delete. Archive or forget the stale row so it stops being recalled.
# Adopt a Skill someone else wrote (/docs/skills/adopt-a-skill-someone-else-wrote)
The point of sharing a Skill is that the second person does not start from a blank page. When somebody on your team perfects a method, you adopt it, adjust the parts that are yours, and run it. You never edit their copy.
## Shared means read only [#shared-means-read-only]
A Skill that has been shared with your department or your whole workspace is frozen at the exact version that was approved. You can open it, read the instructions, see its success criteria and the connections it needs, and copy the instruction text. You cannot edit it in place.
That is on purpose. If shared Skills were editable by everyone, nobody could rely on what they were running. The author can set a shared Skill back to private, edit it, and share it again, but while it is private it disappears for everyone else, so this is a deliberate step and not an accident.
## Adopt it [#adopt-it]
1. Open the shared Skill, either from **Skills** (filter visibility to see what has been shared with you) or from **Marketplace**.
2. Click the add action on the card or in the detail panel.
3. You now have your own private copy in **Skills**, with its own identity.
4. Open your copy and edit anything: the instructions, the success criteria, the name.
Your copy is yours. Editing it never touches the original, and the original being updated later never rewrites your copy.
A Skill carries a method, not access. Adopting a teammate's HubSpot reporting Skill does not give you their HubSpot connection. It runs against your own connected tools, under your own permissions, and produces nothing until you connect what it needs.
## What to change first [#what-to-change-first]
Adopted Skills usually need three edits, in this order:
1. **The connections it assumes.** Read the Skill's connection list. If it expects Notion and your team uses Google Docs, that is an instructions change, not a settings change.
2. **The specifics.** Board names, channel names, document locations, and audiences are almost always the author's, not yours.
3. **The success criteria.** They should describe the outcome you will accept, in your workspace's terms.
## A worked example [#a-worked-example]
Operations shares a `Weekly launch status` Skill that reads its planning page and posts to its team channel. Marketing adopts it, swaps in the campaign planning page, changes the audience and channel, then runs the copy once privately. The original stays unchanged, and Marketing now has a version it can schedule.
## Then schedule it [#then-schedule-it]
Once the copy runs the way you want, click **Create routine** on its detail page to put it on a cadence and pick where the result lands. See [Put a report on a schedule](/docs/routines/put-work-on-a-schedule).
## Share yours back [#share-yours-back]
When your version is better than what you started with, share it so the next person inherits your improvements instead of repeating your first three runs. Depending on how your workspace is set up, a shared Skill may go through a short review before it appears for everyone, and a Skill waiting on that shows as **Needs review**.
## What to check [#what-to-check]
* **The add action is missing or greyed out.** You may already have adopted it. Search **Skills** for the name before adopting again.
* **Your copy fails on the first run.** Read its connection list and connect what is missing, then run it once on demand.
* **You edited the shared original by mistake.** You cannot. If a Skill looks editable, you are looking at your own copy, not the shared source.
* **You want the author's later improvements.** Adoption is a copy, not a subscription. Adopt again to get a fresh copy of the newer version, and keep or delete the old one.
## Related [#related]
# Save a Skill you can reuse (/docs/skills/save-a-skill-you-can-reuse)
A Skill is one method, written down once. It holds the instructions, what it needs to run, what it produces, and how you will know it worked. Once a Skill exists, your Talent can run it on demand, you can pin it to a Routine, and you can share it so a teammate gets your version of the work instead of their own approximation.
Every active member can create and use their own Skills privately. You do not need to be an admin.
## Write one from scratch [#write-one-from-scratch]
1. Open **Skills** in the Workspace group.
2. Click **New skill**, then **Write it yourself**. The same menu also offers **Create with Talent** and **Import a file**.
3. **Name**: what it does, in a few words. `Publish meeting notes`.
4. **Description**: when your Talent should reach for it. `Use when asked to publish reviewed meeting notes`. This line is what decides whether the Skill gets picked, so write it as a trigger, not a summary.
5. **Instructions**: the actual method. Steps, sources, format, tone, what to skip.
6. **Success criteria**: how anyone can tell the run worked. `The page exists in Notion and includes decisions, owners, and a source link.`
7. Click **Create skill**.
You now have a private Skill. Open it, and the detail page shows the instructions, the connections the work needs, its inputs and outputs, and its success criteria.
## Or have your Talent write it [#or-have-your-talent-write-it]
Pick **Create with Talent** from the **New skill** menu, or just ask in web chat, Slack, or Telegram: *"Turn what we just did into a Skill I can reuse."* Your Talent drafts the name, description, instructions, and success criteria, and you edit from there. This is usually faster than a blank form, because the work is fresh.
## Or import one [#or-import-one]
If you already keep methods as Markdown, choose **Import a file** from the **New skill** menu and drop in a Skill folder or a `SKILL.md` file. Packages must be plain text: at most 100 files and 512KB of text in total, no binaries.
Success criteria are worth the extra minute. They are what turns "it produced something" into "it produced the right thing", and they travel with the Skill when a teammate adopts it.
## A worked example [#a-worked-example]
Marketing turns its weekly launch recap into a Skill called `Build launch recap`. The description says when to use it, the instructions name the approved campaign brief and reporting sheet, and the success criteria require five metrics, three learnings, and links to both sources. Run it once on demand. If the result passes that checklist, create a Routine from the Skill.
## Use it [#use-it]
From the Skill's detail page you can run it, or click **Create routine** to put it on a schedule. A Routine created this way pins the exact version of the Skill you are looking at, so later edits never change a running commitment until you decide they should. See [Skill versions and what a Routine pins](/docs/skills/skill-versions-and-what-a-routine-pins).
## Find it again [#find-it-again]
The Skills page has search plus filters for visibility (private, shared with your department, shared with the company), how the Skill runs, and category, and you can sort by name or by last updated. If a Skill is hard to find later, the name is doing too little work: rename it.
## What to check [#what-to-check]
* **Your Talent does not use the Skill when you expect it to.** The description is the trigger. Rewrite it as "Use when asked to..." and name the words people actually say.
* **Runs fail on a missing tool.** Open the Skill and read its connections. A Skill can only use tools you have connected.
* **The import is rejected.** Check the package is text only and inside the 100 file and 512KB limits. Binary and blob backed resources are blocked by design.
* **You cannot edit a Skill you can see.** It is shared, which makes it read only. See [Adopt a Skill someone else wrote](/docs/skills/adopt-a-skill-someone-else-wrote).
## Related [#related]
# Update a Skill without changing a Routine (/docs/skills/skill-versions-and-what-a-routine-pins)
A Skill you are editing is a draft. A Skill version is frozen. Once a version exists it never changes: the instructions, the files, and the contract inside it are fixed, and it carries a content fingerprint so you can tell two versions apart at a glance.
This is the mechanism that makes scheduled work safe. A Routine pins one exact Skill version, so a Monday briefing that has worked for eight weeks keeps running the version that worked, even while you are rewriting the Skill on a Wednesday.
## Freezing a version [#freezing-a-version]
Open a Skill and look at the **Version** field on the detail page.
* **Not frozen** means the Skill exists only as an editable draft. Click **Save as version 1** to freeze it.
* **v1**, **v2**, and so on mean a frozen, immutable version exists. Alongside it you will see the number of files in the package and a short content fingerprint.
Each meaningful save creates the next version. Numbers only ever go up, and an old version is never overwritten or edited, which is why past runs still make sense months later.
## What a Routine pins [#what-a-routine-pins]
When you click **Create routine** from a Skill, the Routine stores the Skill *and* the exact version you were looking at. From then on:
* Editing the Skill and freezing a new version does **not** change the running Routine.
* The Routine keeps its owner, cadence, time zone, delivery target, thread, and history regardless of what happens to the Skill draft.
* To move a Routine onto a newer version, edit the Routine deliberately. Nothing repins itself.
If a Routine's pinned version becomes unavailable, for example because you lost access to a Skill someone shared with you, the Routine stops and shows **Needs attention** rather than quietly running something else. See [Fix a Routine that is not running](/docs/routines/fix-a-routine-that-is-not-running).
## Revising a Skill [#revising-a-skill]
There are three ways to produce a new version, and they all end in the same place:
1. **Edit in place**, then save. The next version is created from your edit.
2. **Ask your Talent** to revise the Skill in chat, then save the result.
3. **Import a revised package**. Export the Skill, edit the folder in your own editor, and use **Import revised version** to bring it back into the same Skill rather than creating a second one.
The import path is why Skills are described as portable. A Skill is a folder of plain text, so it survives outside TalentOS and can be reviewed the way any other document is.
## Sharing freezes things further [#sharing-freezes-things-further]
A shared Skill stays on the exact version your workspace approved. To edit it, the author sets it back to private, revises it, and shares it again. It is unavailable to everyone in between, so plan that change rather than doing it mid week on something people depend on.
## A worked example [#a-worked-example]
Version 2 of your campaign report Skill runs every Friday. On Tuesday you add a new metric and save version 3. Friday's Routine still runs version 2 until you open the Routine and move it to version 3. Test version 3 once before repinning so the scheduled report never changes by surprise.
## What to check [#what-to-check]
* **"Not frozen" and no Create routine button.** Freeze version 1 first. A Routine needs an exact version to pin.
* **You edited a Skill and the Routine did not change.** Working as intended. Edit the Routine to move it onto the new version.
* **Two Skills with the same name.** You probably imported a revision as a new Skill instead of using **Import revised version**. Delete the duplicate and re-import into the original.
* **A revision was rejected.** Packages are capped at 100 files and 512KB of text, and binary files are not accepted.
## Related [#related]
# Find what your Talent did (/docs/settings-notifications/activity-and-receipts)
Chat shows you the conversation. **Activity** shows you the record. Every piece of work your Talent runs, whether you asked in the browser, in a connected messaging channel, or a Routine fired on a schedule, leaves exactly one receipt here.
## Open it [#open-it]
1. Open **Settings** from the **Manage** group.
2. Choose the **Activity** tab.
You see your own runs. Activity is personal history, not a workspace feed.
## The four views [#the-four-views]
* **All**: everything, newest first.
* **Needs you**: only the decisions you can still take, such as an action waiting for your approval. If this is empty, nothing is blocked on you.
* **Completed**: work that finished.
* **Failed**: work that stopped. Start here when something did not arrive.
Narrow further by **time** (today, last 7 days, last 30 days, all) and by **source** (web, connected channels, Routines, actions). There is a search box for the run title and its summary.
## What a receipt tells you [#what-a-receipt-tells-you]
Open a run and you get the narrative first: what was asked, what your Talent did, what it produced. Underneath sits the technical detail, including the governed tool actions it took, the model lane it used, and what the run cost in credits.
Receipts are canonical. If Activity and your memory of a chat thread disagree, the receipt is right.
A run that started as a chat carries a link straight back to that exact conversation, so you can pick the thread up rather than starting over.
## Act on what needs you [#act-on-what-needs-you]
A run waiting on approval shows the decision inline. Approve it and the work continues. You can also allow that kind of action to run automatically from then on, and every automatic run still writes its own receipt, so choosing convenience never costs you the record.
An approval nobody answers eventually expires. It stays in Activity, marked as expired, because a run that stalled is information you need.
## Turn a good run into a Routine [#turn-a-good-run-into-a-routine]
When a successful run contains real tool actions, Activity offers to promote it into a draft Routine you can schedule. The draft is a reviewable starting point rather than an exact replay: receipts do not keep the precise arguments your Talent used, so read the steps before you save it.
## A worked example [#a-worked-example]
A marketing lead's Monday "pull last week's CRM deals into a spreadsheet and post the highlights to team chat" Routine did not post. They open **Settings, Activity**, choose **Failed** and **last 7 days**, and find one run against that Routine with a setup failure: the CRM connection needs reconnecting. They fix the connection, ask their Talent to run it once by hand, then check **Completed** to confirm the spreadsheet and chat post both landed.
## What to check [#what-to-check]
* **Activity is empty.** Check your filters first. A view plus a time range plus a source can easily exclude everything; clear them and start from All.
* **You cannot find a teammate's run.** Activity is per person. Ask them to open theirs.
* **A run shows as failed with no obvious reason.** Look for a setup failure on its Routine. A disconnected or re-authorised connector is the most common cause.
* **A Routine has no runs at all.** It may never have fired. Check the Routine's schedule, and check the workspace is not read only. See [Fix a read-only workspace](/docs/billing/read-only-mode).
* **Nothing loads.** Each section reports its own error. Reload once, then contact support with the message shown.
# Emails from TalentOS (/docs/settings-notifications/emails-from-talentos)
TalentOS is deliberately quiet by email. Almost everything it has to tell you arrives in the app, in the bell, or in the channel you asked from. Set your expectations accordingly, especially if you are waiting on something.
## What does arrive by email [#what-does-arrive-by-email]
**Sign-in and invitation emails.** Workspace invitations, address verification and password resets are sent by the secure sign-in provider that handles your account identity. These are the emails most people are actually waiting for.
## What does not [#what-does-not]
There is no per notification email preferences screen, and no digest. If you want to be told when work finishes, use the browser notification toggle or a chat channel. See [Get notified when work finishes](/docs/settings-notifications/get-notified).
Do not build a work process around an email arriving. If a teammate needs to know a Routine produced something, send the result to a connected messaging channel or check Activity.
## A worked example [#a-worked-example]
An office manager invites five new starters on a Friday. Two of them never see the invite. Rather than resending blindly, she checks the pending invitations list in Team, confirms both addresses are spelled correctly, asks the two to search their spam folder for the sender, and has them add it to their contacts. Both invites are sitting in junk. Once accepted, the workspace seat count rises from eight to thirteen on its own.
## What to check [#what-to-check]
1. **Check spam or junk.** This is the answer most of the time. Move the message to the inbox and mark the sender as trusted so the next one lands.
2. **Check the address.** Emails go to the address on the account. A typo in an invite means the invite went somewhere else; ask an admin to re-send to the corrected address.
3. **Check corporate filtering.** Some workspaces quarantine mail from new senders. Your IT team can allow the sender.
4. **Do not wait on it if there is another route.** An invited teammate can be re-invited. Send work notifications to a connected messaging channel rather than chasing email.
5. **Still nothing?** Contact support with the exact address you expected it to reach.
## Related [#related]
# Find the right settings tab (/docs/settings-notifications/find-your-settings)
Settings holds more than preferences. Your run history, your credit balance, your Talent's configuration and your workspace's API access all live here. This page tells you which tab to open.
Open **Settings** from the **Manage** group at the bottom of the left nav.
## If you are a member or manager [#if-you-are-a-member-or-manager]
| Tab | Use it to |
| -------- | -------------------------------------------------------------------------- |
| Profile | Change your display name, and pick light, dark or system appearance |
| Account | See your sign-in method, change your password, sign out |
| Talent | Configure your Talent: its model lane, memory, connections and permissions |
| Activity | Read the durable history of everything your Talent has run |
| Billing | See the workspace plan and the shared credit pool, read only |
| Legal | Privacy Policy and Terms |
## If you are an admin or the owner [#if-you-are-an-admin-or-the-owner]
| Tab | Use it to |
| --------------- | -------------------------------------------------------------------------------------------- |
| General | Rename the workspace, jump to Team and departments |
| API & MCP | Create and revoke API keys, and connect MCP clients |
| Usage | See 30 days of credit spend by member and by Routine |
| Talent | Configure your own Talent |
| Activity | Read the durable run history |
| Plans & Billing | Start, change or cancel a plan, update the card, buy top-ups |
| Support & Legal | Email support, report a bug, suggest a feature, join the community, read the legal documents |
Admins have both surfaces. Your own profile, appearance and account live on the member Settings page; workspace-level controls live on the admin one.
## Three things people look for in the wrong place [#three-things-people-look-for-in-the-wrong-place]
**Run history is not next to chat.** It is the **Activity** tab in Settings. Chat is for doing the work; Activity is the durable record of what was done. See [Find what your Talent did](/docs/settings-notifications/activity-and-receipts).
**Notification choices are not a single preferences screen.** They are attached to the things that produce them: a per device toggle in the chat header, the bell in the top bar, and your connected messaging channel. See [Get notified when work finishes](/docs/settings-notifications/get-notified).
**Inviting people is not in Settings directly.** Use **Team** in the Manage group, or the Team and departments link from the admin General tab.
## What you cannot change here [#what-you-cannot-change-here]
Your **email**, your **department** and your **role** appear on the Profile tab but are read only, because the workspace owns them. An admin changes your department and role from Team. Your email is your sign-in identity.
## A worked example [#a-worked-example]
A new ops manager wants to know why last Tuesday's client report never landed in team chat. They open **Settings**, choose **Activity**, filter to **Failed** and the last **7 days**, and find the Routine run that stopped waiting for an approval nobody gave. Two tabs later, in **Plans & Billing**, they confirm the workspace still has credits, so the fix is the approval, not the bill.
## What to check [#what-to-check]
* **A tab you expect is missing.** Usage appears only for admins and owners. Check your role in Team, then reload.
* **You are an admin and cannot change the plan.** Only the workspace owner can. See [Change plan, update your card, or cancel](/docs/billing/manage-your-subscription).
* **Settings will not load.** Reload once. If it still fails, the page shows the reason; contact support with that message.
# Get notified when work finishes (/docs/settings-notifications/get-notified)
Long jobs are the point of having a Talent, and staring at a spinner is not. TalentOS gives you three different signals, each for a different situation.
## Ping this device when a turn finishes [#ping-this-device-when-a-turn-finishes]
The bell icon in the **chat header** opts **this browser on this device** into a notification when your Talent finishes what it is doing.
1. Open your Talent in the browser.
2. Click the bell in the chat header.
3. Accept your browser's permission prompt when it appears.
4. The icon changes to show notifications are on.
Ask for something slow, switch to another tab, and you get a system notification the moment it is done.
This is **per device and per browser**. Turning it on at your desk does not turn it on for your laptop or your phone. Turn it on again wherever you want to be pinged.
## The bell in the top bar [#the-bell-in-the-top-bar]
The bell in the app header is a **bounded inbox for things that need a decision**, not a history feed. It surfaces items such as:
* an action your Talent is waiting for you to approve
* a Skill or Routine version submitted for your review, or changes requested on yours
* comments addressed to you
* "Private login needed" when your Talent hit something only you can sign in to
* "Memory is full" when your active memory pages have hit their ceiling
* "Workspace credits are running low"
Clear it by acting on the items, not by scrolling it. For the durable record of everything that ran, use [Activity](/docs/settings-notifications/activity-and-receipts) instead.
## Let the work come to you in a messaging channel [#let-the-work-come-to-you-in-a-messaging-channel]
If you would rather not watch the browser at all, ask for the work in the channel you already live in. Your Talent answers where you asked, so a Routine set up from team chat reports back there, and quick approvals can be answered with a tap without opening the console.
## A worked example [#a-worked-example]
An agency producer asks their Talent for a competitive teardown that will take several minutes, clicks the chat header bell, accepts the browser prompt, and goes back to email. Six minutes later the notification arrives and the Artifact is waiting. The next day they set the same job up as a weekly Routine and point it at team chat, so Monday's teardown arrives without anyone watching the app.
## What to check [#what-to-check]
* **The bell is not in the chat header.** Your browser does not support web push. Use a connected messaging channel instead.
* **The bell shows as blocked.** You previously denied notifications for this site. Only your browser settings can undo that; re-allow notifications for the site, then click the bell again.
* **You allowed notifications but nothing arrives.** Check that your operating system is not in a focus or do-not-disturb mode, and confirm you enabled it on the device you are actually watching.
* **You are getting nothing anywhere.** Confirm the work actually ran. Open **Settings, Activity** and look at Failed.
* **The header bell never empties.** Some items need a decision from you and stay until you make it. Expired approvals stay visible on purpose, because a stalled run is something you need to know about.
# Update your name, appearance and sign-in (/docs/settings-notifications/your-profile-and-account)
Your personal details live on two tabs in **Settings**: **Profile** for how you appear and how the app looks, and **Account** for how you sign in.
## Change your display name [#change-your-display-name]
1. Open **Settings** from the **Manage** group.
2. Stay on the **Profile** tab.
3. Edit **Full name** and save.
This is the name your teammates see across the workspace. It refreshes everywhere once your account finishes syncing, usually within seconds.
## Switch between light and dark [#switch-between-light-and-dark]
On the same **Profile** tab, the Appearance card offers three choices:
* **Light**: always light.
* **Dark**: always dark.
* **System**: follows your device, so it flips with your operating system's own light and dark schedule.
The choice applies to you alone, not to your workspace.
## What you cannot change here [#what-you-cannot-change-here]
Your **email**, **department** and **role** are shown on Profile but are read only.
* **Email** is your sign-in identity.
* **Department** and **role** are workspace managed. An admin changes them from **Team**. If you have not been put in a department yet, this reads **Not assigned**.
Ask an admin rather than looking for a control that is not there.
## Review your sign-in and change your password [#review-your-sign-in-and-change-your-password]
Open the **Account** tab. It shows:
* **Sign in method**, along with the email tied to your account.
* **Change password**, which opens your secure account profile. This is only available when your account actually uses a password. If you use single sign-on, the row is disabled, and you change that password with the sign-in service instead.
* **Sign out**, which asks you to confirm and then returns you to the sign-in page.
Account security is handled by the secure sign-in provider rather than by a form inside TalentOS, which is why the password flow opens its own panel.
## A worked example [#a-worked-example]
A new designer joins mid week with single sign-on and appears to everyone as "dh". She opens **Settings, Profile**, sets her full name to "Dana Hale", and picks **Dark** appearance. She notices her department reads **Not assigned**, messages her admin, and by the afternoon her Profile shows Brand, which also puts her department's shared memory in reach of her Talent.
## What to check [#what-to-check]
* **Change password is greyed out.** Your account does not use a password. Change it with the provider you sign in through.
* **Your new name has not appeared for teammates.** Give account sync a moment and reload. If it is still stale after a few minutes, contact support.
* **Department still says Not assigned.** Only an admin can set it, from Team.
* **The theme keeps reverting.** You are probably on **System**, and your device is switching. Choose Light or Dark explicitly.
* **Settings will not load at all.** Reload once, then contact support with the message the page shows.
# Choose a cadence and where results land (/docs/routines/choose-a-cadence-and-where-results-land)
Two settings decide whether a Routine is useful: when it fires, and where the result shows up. Both live in the Routine form, and both read back in plain language so you can check them before you save.
## The four cadences [#the-four-cadences]
Under **When**, pick one:
* **Once** runs a single time. You choose a date, a time, and a time zone: *Once on 14 Oct at 9:00 AM*.
* **Hourly** runs every hour at a minute you choose: *Every hour at :15*.
* **Daily** runs once a day: *Every day at 8:00 AM*.
* **Weekly** runs on the days you select: *Every Mon, Wed & Fri at 9:00 AM*.
There is no cron field, and cron text is rejected on purpose. If you need a shape the four cadences cannot express, run two Routines.
## Time zone [#time-zone]
Every recurring Routine stores an explicit time zone, pre-filled from your browser. It is part of the schedule sentence, so check it when you travel or when you are scheduling for a team in another region. A 09:00 Monday briefing set in one zone does not follow you to another.
## Start and end dates [#start-and-end-dates]
For anything other than **Once**, you can set an optional **Starts** and **Ends** date. Use **Starts** to arm something ahead of a launch, and **Ends** to close a commitment that has a natural finish, like a campaign or a quarter.
Every Routine has an end. If you do not set one, TalentOS supplies a bound for you, so nothing fires forever after the reason for it has gone. A Routine that reaches its end shows as **Completed** rather than silently disappearing.
## Where the result lands [#where-the-result-lands]
**Result delivery** is where the complete result is sent. Tick as many as you want:
* **TalentOS** puts it in your Talent home in the web console.
* **Slack** sends it to you as a direct message, or to a channel you pick from the list.
* **Telegram** sends it to your connected Telegram chat.
* **iMessage** sends it to the number you paired.
A channel only appears once you have connected it, so if Telegram is missing from the list, connect Telegram first and reopen the form.
Delivery is not history. Ticking nothing is allowed: the run still happens and the result is still saved to the Routine, but nothing is pushed to you and you will only see it if you go looking. Pick at least one target for anything you actually need to read.
## Asking for a schedule in chat [#asking-for-a-schedule-in-chat]
When you set a Routine up by talking to your Talent, an incomplete time is treated as a question, not an error:
* **"Every Monday"** with no time resolves to a default time. It recurs, so a wrong guess corrects itself next week and the schedule is read back to you either way.
* **"Tomorrow morning"** resolves inside that window. The exact minute is a correction, not a guess.
* **"Monday"** for a one-off run asks you which time you meant, and offers candidates. It fires once, so guessing wrong would be the whole outcome.
* **"Tomorrow"** asked between midnight and 04:00 always asks which date you meant, and offers both.
Genuinely unparseable text such as "sometime soonish", an unknown time zone, and a time already in the past are refused outright.
## A worked example [#a-worked-example]
Your five person marketing team needs a campaign pulse every weekday at 4:30 PM in the team's local time. Create one **Weekly** Routine, select Monday through Friday, confirm the time zone, and choose the team Slack channel plus TalentOS for delivery. Set an end date for the day after the campaign closes. The schedule sentence and delivery list are your final preflight check.
## What to check [#what-to-check]
* The schedule **sentence** matches what you meant. It is what gets stored, not the words in your instructions.
* The **time zone** is the one the audience lives in.
* At least one **delivery** target is ticked.
* If a Slack channel you picked earlier has vanished from the list, use **Refresh** next to the channel picker. The available channel list expires and is re-fetched.
## Related [#related]
# Fix a Routine that is not running (/docs/routines/fix-a-routine-that-is-not-running)
Schedules fire every minute, so a Routine that misses its slot has a reason, and TalentOS records it. Work through this list in order and you will find it.
## 1. Read the badge [#1-read-the-badge]
Open **Routines** and look at the card.
* **Needs attention** means TalentOS stopped the Routine rather than run it badly. The card shows the reason in plain language, and usually a link straight to the fix.
* **Paused** means a person stopped it. Open the Routine and click **Resume Routine**.
* **Completed** means it reached its end date. Create a new Routine, or edit this one and extend the end date, then resume it.
* **Active with no next run** almost always means the end date has passed.
## 2. Reconnect the tool [#2-reconnect-the-tool]
The most common cause of **Needs attention** is a connection that expired or was revoked. A Routine runs with your own connected tools, and most connections need re-authorising eventually. When the Monday HubSpot summary stops arriving, HubSpot is the first thing to check.
1. Open the Routine and read the **Connections** panel. It lists exactly which connections this Routine's work needs.
2. Follow the fix link on the card, or open **Connections** in the Manage group.
3. Reconnect the tool. Sign in again if you are prompted.
4. Return to the Routine and click **Resume Routine**.
Reconnecting on its own does not restart the Routine. TalentOS will not silently resume a commitment it stopped, so you have to resume it yourself. Check the card says **Active** with a next run time before you walk away.
## 3. Answer the question your Talent asked [#3-answer-the-question-your-talent-asked]
If you set the Routine up by chatting and it never appeared, look in that conversation. An incomplete schedule comes back as a short question rather than a refusal, and the Routine is not created until you answer it.
You will see this when you gave a one-off day with no time ("do this Monday"), or when you said "tomorrow" between midnight and 04:00. Pick one of the offered options and the Routine is created.
You will get a flat refusal, not a question, for cron syntax, an unknown time zone, a time already in the past, or text that carries no time at all. Rephrase using a plain sentence like "every Tuesday at 4pm".
## 4. Confirm whether it ran but did not reach you [#4-confirm-whether-it-ran-but-did-not-reach-you]
A silent Routine is not always a stopped one.
1. Open the Routine and click **Open Routine history**. If the run is there, it ran.
2. Check **Result delivery** in the Routine. If nothing is ticked, results are saved but never pushed.
3. If Slack was ticked and the channel is gone from the picker, use **Refresh** beside it and select the channel again. Channel lists expire.
4. For a definitive answer, open **Activity** in **Settings** and find the run's receipt.
## 5. Check the instructions, not the schedule [#5-check-the-instructions-not-the-schedule]
If runs land on time but the output is wrong, the schedule is fine and the brief is not. Open **Edit routine** and make the instructions specific: name the source, name the fields, say the length, say what to exclude. Then run it once on demand and read the result before waiting another week.
## A worked example [#a-worked-example]
A Monday HubSpot summary shows **Needs attention** and did not reach Slack. Open the Routine and follow its connection warning to reconnect HubSpot. Return to the Routine, choose **Resume Routine**, and run it once on demand. If history shows a successful run but Slack is still empty, refresh **Result delivery**, select the Slack destination again, and confirm the delivery in Activity before relying on next Monday's schedule.
## What to check [#what-to-check]
* A Routine pinned to a Skill you no longer have access to will stop. Ask the person who shared the Skill to share it again.
* If the reason on the card mentions setup, open the Routine's source and follow the setup link before resuming.
* If the same Routine blocks repeatedly on the same connection, that connection is being revoked somewhere upstream. Reconnect it, then check whether an admin policy or a password change is expiring it.
## Related [#related]
# Put a report on a schedule (/docs/routines/put-work-on-a-schedule)
A Routine is work your Talent runs without being asked. You write the instructions once, pick when it runs and where the result lands, and it keeps running. Nobody has to remember Monday morning.
This page walks one real example end to end: a Monday 09:00 pipeline summary built from HubSpot and delivered to a Slack channel.
## A worked example [#a-worked-example]
1. Open **Routines** in the Workspace group of the left navigation.
2. Click **New routine**, then **Set up manually**. (The other option, **Create with Talent**, is covered below.)
3. **Name** it something a teammate would recognise: `Monday pipeline summary`.
4. **Category** groups it on the Routines page. Pick the one that fits.
5. **Instructions** is the whole brief. Write it the way you would brief a person:
> Pull every open HubSpot deal closing in the next 30 days. Group them by stage. Give me total value per stage, then list any deal that has not moved in 14 days. Keep it under 200 words.
6. Under **When**, choose **Weekly**, click the day chips to select Monday, set the time to 09:00, and confirm the time zone. The schedule reads back as a sentence: *Every Monday at 9:00 AM*. Check that sentence before you continue.
7. Under **Result delivery**, tick **Slack** and choose the channel. Delivery is where the finished result is sent, and you can pick more than one place.
8. Click **Create**.
You land on the Routine's detail page. It shows what your Talent will do, the schedule, the connections the work needs, and its run history.
## Or just ask [#or-just-ask]
You do not have to use the form. In web chat, Slack, or Telegram, tell your Talent what you want:
> Every Monday at 9am, summarise our HubSpot pipeline and post it to #revenue.
Your Talent creates the Routine and reads the schedule back to you. If you leave something out that matters, it asks a short follow up question rather than refusing. See [Choose a cadence and where results land](/docs/routines/choose-a-cadence-and-where-results-land).
A Routine runs with **your** connected tools and your permissions. The HubSpot data in the example is the data your own HubSpot connection can see. If you have not connected HubSpot yet, connect it before the first run.
## Change it later [#change-it-later]
Open the Routine and use **Edit routine** to change the name, instructions, schedule, or delivery. **Pause Routine** stops future runs and keeps everything else intact. **Resume Routine** starts it again from the next scheduled slot.
## What to check [#what-to-check]
* **The badge says "Needs attention".** Something stopped it, most often a connection that expired. Open the Routine and read the message on the card, then see [Fix a Routine that is not running](/docs/routines/fix-a-routine-that-is-not-running).
* **Nothing arrived in Slack.** Check **Result delivery** on the Routine. If no delivery target is ticked, the run still happens and the result is saved to the Routine's history, but nothing is pushed to you.
* **The schedule sentence is wrong.** The sentence under **When** is what actually gets stored. If it does not match what you meant, fix it there rather than in the instructions.
* **The result is thin or off target.** The instructions are the brief. Say what to include, what to leave out, and how long it should be, then run it once from the Routine page to check.
## Related [#related]
# See what a Routine actually did (/docs/routines/run-history-and-routine-states)
A Routine that runs unattended is only trustworthy if you can check its work. Every run is kept, whether or not it was delivered anywhere, and every Routine has exactly one conversation thread that all of its runs write into.
## One Routine, one thread [#one-routine-one-thread]
Each Routine owns a single persistent thread. Monday's run and next Monday's run land in the same conversation, in order. That is deliberate: you can scroll a quarter of pipeline summaries in one place, and you can reply in the thread to steer the next run without recreating anything.
To open it, go to **Routines**, open the Routine, and in the **Routine history** panel click **Open Routine history**. It opens the exact conversation the runs are writing to. On the console rail, the same threads are grouped under **Routine chats**.
## What the states mean [#what-the-states-mean]
The badge on a Routine card tells you where it stands:
* **Active** is scheduled and will fire at the next slot. The card shows the next run time.
* **Paused** means you stopped it. Nothing fires until you resume. Use **Pause Routine** and **Resume Routine** on the detail page.
* **Needs attention** means TalentOS stopped it because it could not run safely, most often an expired connection. The card carries the reason and usually a link that fixes it.
* **Held** means a run is being withheld for now rather than cancelled.
* **Completed** means the Routine reached its end date or its end condition. Its history stays readable.
A Routine in any state other than Active shows no next run time, so an empty **Next run** is a signal, not a display bug.
## Receipts are the record [#receipts-are-the-record]
The Routine thread shows the work. The **receipt** is the governed record of it: one receipt per run, listing what your Talent actually did, which tools it used, and what it produced. Receipts live in **Activity**, inside **Settings**. Go there when you need to answer "did this really send" or "what did it touch", rather than reading back the conversation.
Delivery, history, and receipts are three different things. Delivery is a push to Slack, Telegram, iMessage, or your Talent home. History is the Routine's own thread. The receipt is the durable record in Activity. A missing Slack message does not mean the run did not happen: check history and Activity before assuming a failure.
## Run it now [#run-it-now]
You do not have to wait for the next slot. Open the Routine and run it from the detail page when you want to test a change to the instructions, or when you need this week's output early. An on demand run writes to the same thread and produces the same kind of receipt as a scheduled one.
## Editing without losing the thread [#editing-without-losing-the-thread]
**Edit routine** changes the name, category, instructions, schedule, and delivery in place. The thread and the run history stay attached. You do not need to delete and rebuild a Routine to change how it works, and doing that would cost you the history.
**Delete Routine** removes the standing commitment. Do it when the work is genuinely over, not to reset a bad run.
## A worked example [#a-worked-example]
Your Monday pipeline Routine shows **Active**, but no message reached the team channel. Open **Routine history** first. If the run and summary are there, open its receipt in **Activity** and check the delivery step. A failed delivery means the scheduled work ran and only the destination needs attention. An empty history means you should check the next run time, time zone, and end date.
## What to check [#what-to-check]
* **No next run time on an Active Routine.** Reopen the schedule and confirm the end date has not already passed.
* **Runs happen but you never see them.** No delivery target is ticked. Add one in **Result delivery**.
* **The thread is empty but the badge says Active.** It has not fired yet. Confirm the time zone in the schedule sentence.
* **You need proof for someone else.** Use the receipt in **Activity** under **Settings**, not a screenshot of the chat.
## Related [#related]
# Ask your Talent for real work (/docs/talent/ask-your-talent)
You have your own Talent. It is an AI teammate that does work, not a search box. This page gets you from an empty composer to a finished piece of work in one sitting.
## What a Talent is [#what-a-talent-is]
Every member of your workspace gets one Talent of their own. Yours has its own memory of how you work, its own response style, and its own set of connected apps. Your teammates each have theirs. Nobody shares one.
The important part: **your Talent acts with your access and only yours**. If you connected Gmail, it reads your inbox. If a teammate connected HubSpot and you did not, your Talent cannot touch HubSpot. There is no way for it to borrow a colleague's account.
## Do this first [#do-this-first]
1. Open **Home**. That is the front door, and the composer sits in the middle of it.
2. Go to **Connections** and connect at least one app you actually work in. Gmail, Google Sheets, Notion, Slack, HubSpot, Asana, Airtable, GitHub, Jira, Linear, Outlook, Google Docs, Google Drive, and Google Calendar are all available today.
3. Come back to Home and describe the job in plain language. Say what you want at the end, not which buttons to press.
## A worked example [#a-worked-example]
Type this into the Home composer, with Gmail and Google Sheets connected:
> Go through my inbox from the last seven days, find every email that mentions a contract renewal or an expiring plan, and build me a Google Sheet called "Renewals watchlist" with the company, the contact, the date they mentioned, and a one line summary.
Your Talent reads the inbox with your Gmail connection, works out what counts, and creates the sheet in your Drive. Writing the sheet is a governed action, so depending on your permissions it either goes ahead and tells you, or pauses and asks you to approve it first. Approve it and the work continues from exactly where it stopped.
You get back a short reply and a link. Every run also leaves a receipt you can open later.
## What it can do without any connector [#what-it-can-do-without-any-connector]
Even before you connect anything, your Talent can search the web under governance, run code to do real calculation, and produce artifacts you can download as DOCX, PDF, PPTX, or XLSX files. "Turn these numbers into a board slide" works on day one.
Ask for the outcome, not the method. "Summarize last week's support email into three themes with examples" beats "search my inbox for the word urgent".
## If you want it to happen every week [#if-you-want-it-to-happen-every-week]
Anything you ask for once can become a **Routine**, which is the same work on a schedule or on demand. Ask your Talent to make the job repeatable and it will draft one for you.
## Where to go next [#where-to-go-next]
## What to check [#what-to-check]
* **It says it cannot see an app.** The app is not connected for you, even if a teammate connected it. Open Connections and connect it under your own account.
* **Nothing happens after you hit send.** Check the Home **Today** section. Work that needs a decision from you waits there and in the header bell.
* **It asked a question instead of doing the work.** That is deliberate when a request is ambiguous in a way that would waste a run. Answer with the one missing detail and it continues.
* **The result is thinner than you expected.** Say so in the same conversation. It keeps the context and revises rather than starting over.
# Choose your model lane and response style (/docs/talent/model-and-response-style)
Two settings change how your Talent feels day to day: which model lane it runs on, and how it writes. Both are yours to set, and neither is gated behind a plan.
## Pick a model lane [#pick-a-model-lane]
1. Open **Home** and go to **Manage**, then **Model**. Admins can also reach it from the usage dashboard in Settings.
2. Read the list. Lanes are listed cheapest first, and each shows a five pip cost gauge.
3. Pick one and save. It applies to your Talent from the next turn.
**Every lane is available on every plan.** Team, Power, and Max all see the same list. The difference between lanes is what a turn costs in credits and how much judgment you get, not which plan you bought.
The five pips are the quick signal: fewer pips means less credit burn per turn. The exact credit multiplier is in the tooltip if you want the number. One lane is marked **Default**, and if you never touch this setting that is what you are on.
Changing lanes does not change what your Talent can do. It has the same connected apps, the same memory, and the same permissions on every lane. Only cost and judgment move.
### Which lane to pick [#which-lane-to-pick]
* **Everyday work, high volume.** Stay on the default or drop to a cheaper lane. Pulling numbers, sorting an inbox, and filling a sheet do not need a premium lane.
* **Work you will send to a customer or a board.** Move up a lane. Drafting, analysis, and multi step tool use are where a stronger lane earns its credits.
* **Long research runs.** A long context lane handles a big pile of source material better than a fast one.
A practical pattern: run your Routines on a cheap lane, and switch up when you are drafting something that matters.
## A worked example [#a-worked-example]
Your operations team runs a daily inbox summary and a monthly board update. Keep the daily Routine on the default lane and set its response style to **Concise**. Before drafting the board update, switch to a stronger lane and ask for a structured narrative with risks and evidence. The Routine keeps its connections and permissions, while the one-off board task gets more judgment.
## Set the response style [#set-the-response-style]
In **Manage**, pick one of five styles:
* **Balanced** is the default. Clear, warm, and practical.
* **Concise** gives short answers with no preamble, often one or two sentences.
* **Coach** is patient, breaks work into small steps, and checks in as it goes.
* **Creative** offers a few angles instead of narrowing to one answer too early.
* **Custom** lets you write your own instructions in plain language.
## What Custom instructions can and cannot do [#what-custom-instructions-can-and-cannot-do]
Custom instructions change **tone, structure, and working style**. That is their whole job.
They cannot widen what your Talent is allowed to see, add a capability it does not have, or skip an approval. This is enforced in the product, not just requested politely, so writing "you may skip approvals" in your instructions does nothing. If you want your Talent to act more freely, change **Permissions** instead.
## What to check [#what-to-check]
* **The model list will not load.** The picker checks lane availability before letting you switch. Wait a moment and reopen it. Your Talent keeps running on its current lane.
* **Replies still feel long after choosing Concise.** Style shapes voice, not content. If the work genuinely needs detail you will still get detail. Ask for a specific length in the request itself.
* **Credits are going faster than expected.** Check which lane you are on first. A premium lane on a daily Routine is the usual cause. An admin can see where workspace credits are going on the usage dashboard in Settings.
* **You changed the lane and nothing seems different.** It applies from the next turn, not retroactively to a run already in flight.
# Let your Talent nudge you, and turn it down (/docs/talent/nudges-and-notifications)
Your Talent does not only answer. It can notice something in your connected mail or calendar and message you first. You control how often that happens and where it lands.
## Set your proactivity pace [#set-your-proactivity-pace]
Open **Home**, go to **Manage**, then the proactivity setting. You get three controls:
1. **On or off.** Off means your Talent never starts a conversation. It still answers you normally.
2. **Reach me via.** Which channel a nudge arrives on.
3. **Pace.** How many nudges a day you are willing to accept.
The pace is a hard daily cap, counted per person per local calendar day:
| Pace | Nudges per day |
| ---------------- | -------------- |
| Off | none |
| Quiet | 1 |
| Normal (default) | 2 |
| Eager | 5 |
These are ceilings, not targets. On a quiet week you will get fewer, or none.
Quiet hours are built in. Between 10pm and 7am in your local time, your Talent does not send proactive messages at all, whatever your pace.
## What it actually watches [#what-it-actually-watches]
Nudges come from what you have connected, not from guesswork. Your Talent watches the head of your connected mailboxes and your connected calendars, up to five of each, and looks for things you committed to that are about to matter.
It also will not nudge you twice about the same thing. A given subject is deduplicated for three days, so a stubborn unanswered email does not become a daily complaint.
## A worked example [#a-worked-example]
You are on Normal pace with Telegram as your channel. Google Calendar and Gmail are connected. On Wednesday morning you get one message:
> You have the Acme QBR at 2pm and their last email asked for updated usage numbers, which nobody has sent. Want me to pull them and draft the reply?
You answer "yes". That is your one useful interruption for the day. Nothing else arrives.
## Get pinged when a long run finishes [#get-pinged-when-a-long-run-finishes]
Separately from nudges, you can ask the browser to notify you when a turn is done. In the chat header there is a bell. Click it, allow notifications when your browser asks, and that device gets a push when a run finishes.
This is **per device and per browser**. Turning it on at your desk does not turn it on for your laptop.
## The header bell is not history [#the-header-bell-is-not-history]
The bell in the console header is a bounded inbox of things that still need you: approvals waiting, requested changes, comments on work you shared, and low credit warnings. Items leave it when they are resolved.
It is not a log. For the full record of what ran, use [Activity](/docs/talent/receipts-and-activity) in Settings.
## What to check [#what-to-check]
* **You get no nudges at all.** Check the pace is not Off, check the chosen channel is still connected, and remember that quiet hours cover 10pm to 7am locally.
* **You get too many.** Drop from Eager to Normal or Quiet. If a specific topic keeps coming back, tell your Talent it is handled and it will stop treating it as open.
* **The bell button in the chat header is missing.** Your browser does not support push notifications. Safari on older versions and some in app browsers are the usual cases.
* **The bell shows as blocked.** You previously denied notifications for this site. TalentOS cannot re-ask. Allow notifications for the site in your browser settings, then toggle it again.
* **Notifications work on one machine but not another.** That is expected. Turn it on again on the second device.
# Check what your Talent actually did (/docs/talent/receipts-and-activity)
When your Talent does work, you should not have to take its word for it. Every run leaves one receipt, and Activity is where you read them.
## Find a run [#find-a-run]
1. Open **Settings**.
2. Go to **Activity**.
3. Pick a view: **Needs you**, **Completed**, or **Failed**.
**Needs you** contains only decisions you can still take, so an empty Needs you means nothing is blocked. Completed and Failed cover your full history, not just what happens to be cached in your browser.
## What a receipt tells you [#what-a-receipt-tells-you]
A run is the story container, and the receipt is the evidence. Open one and you can see:
* What the run was asked to do, and how it finished.
* Which governed actions it took, such as sending an email or writing to a sheet, and whether each was allowed outright or approved by you.
* Which model lane it used and what it cost.
* When each part happened.
If a workspace connection was used to cover for a connector you had not connected yourself, the receipt names the colleague who owns that connection and the app it belongs to. It never shows their credentials.
For work delivered to a private iMessage chat, the receipt says **Private iMessage chat** plus the delivery status. It never shows a phone number.
## A worked example [#a-worked-example]
Your Monday pipeline Routine ran while you were asleep and the Slack message never arrived. In Activity you open **Failed**, find Monday's run, and the receipt shows the HubSpot read succeeded, the summary was written, and the Slack delivery failed. That tells you the problem is the channel, not the work. Reconnect Slack in Connections and rerun. You did not have to guess.
## Turn a good run into a Routine [#turn-a-good-run-into-a-routine]
When a successful receipt contains at least one governed action worth repeating, Activity offers to promote it into a Routine. Take the offer when the same job will come round again.
A promoted Routine is a reviewable starting point, not an exact replay. Receipts do not keep the exact arguments of every tool call, so read the draft before you schedule it.
## Activity versus the bell [#activity-versus-the-bell]
These two are easy to confuse, and mixing them up is the usual reason people think something is missing.
* **The header bell** is bounded attention. It holds only unresolved things that need you now, and items disappear when they are resolved.
* **Activity** is the durable record. Everything that ran is there, resolved or not.
If something vanished from the bell, it was not deleted. It was resolved. Look for it in Activity.
## What to check [#what-to-check]
* **A run shows no cost or model.** Provider telemetry occasionally does not come back. The outcome of the run is still accurate, the technical detail is just incomplete.
* **You cannot find a teammate's run.** You cannot see it, and neither can an admin. Runs are private to the person who started them. See [who can see your chats and memory](/docs/talent/who-can-see-your-work).
* **The promote option is not offered.** The run either failed, or it did nothing that can be turned into a repeatable step. Ask your Talent to build the Routine from scratch instead.
* **A receipt is missing for something you know ran.** Check you are on the right view. A run that is still going lives in Needs you or in its own conversation until it finishes.
# Reach your Talent from Slack, Telegram, or iMessage (/docs/talent/slack-telegram-imessage)
You do not have to sit in a browser tab to use your Talent. It answers in Slack, Telegram, and iMessage, and it can start a conversation there when something needs you.
## What each surface is for [#what-each-surface-is-for]
Each surface has a job. Using the right one saves you a lot of friction.
* **The web console** is for understanding and governance. Read what happened, change what your Talent may do, edit memory, manage connections, open artifacts.
* **Slack and Telegram** are for attention and quick decisions. "This needs you", plus a fast yes or no on an approval. Anything deeper links back into the console.
* **iMessage** is the same as Slack and Telegram, on your phone, with voice notes and tapbacks.
## Connect a channel [#connect-a-channel]
1. Open **Connections**.
2. Under channels, choose **Telegram** or **iMessage** and follow the pairing steps.
3. For Slack, the TalentOS Slack app is installed once for the whole workspace by an admin. Once it is in, you can direct message your Talent there straight away.
4. Send a test message. Your Talent replies in the same place.
Telegram has no plan gate. iMessage is open to every workspace, including trials.
## A worked example [#a-worked-example]
You are on a train. Your Talent messages your Telegram: "Your Monday pipeline note is ready, and one deal moved to closed lost. Want me to draft a note to the account owner?" You reply "yes, keep it short". It drafts the reply in Gmail and tells you it is sitting in your drafts. You approve nothing else and open nothing. Total time, forty seconds.
## In a shared Slack channel [#in-a-shared-slack-channel]
Your Talent behaves differently in a channel than in your DM, and this is deliberate.
* **It stays discreet.** In a channel other people can read, it never volunteers your memories, notes, documents, schedules, or anything it read from your connected accounts. If the honest answer needs private detail, it keeps the channel reply brief and moves the rest to your DM.
* **It can tag teammates** in the channel it is replying in, so the right person gets notified.
* **It cannot direct message anyone but you.** If you ask it to message a colleague privately, it will say so plainly and leave that to you.
There are two different Slack things. The TalentOS Slack app is how you talk to your Talent. The personal Slack connector is how your Talent reads and searches Slack using your own account. Connecting one does not connect the other.
## iMessage, exactly [#imessage-exactly]
iMessage runs over a **shared TalentOS line**, not a phone number of your own. That has real consequences worth knowing up front:
* Pairing is rate limited: at most **40 new pairings per line per day**, and at most **10 per workspace per day**. If a whole team pairs at once, some people will be asked to try again tomorrow.
* iMessage is one to one with you. Group chats are not supported.
* You can react to your Talent's messages with a tapback, and it can put one on yours.
* Voice notes work here too. See [talking to your Talent with voice notes](/docs/talent/voice-notes).
## What to check [#what-to-check]
* **Your Talent says Slack is not connected while replying in Slack.** It is talking about your personal Slack connector, not the chat. Chat still works. Connect the personal account in Connections if you want it to search other channels.
* **iMessage shows as temporarily unavailable in Connections.** The shared line is down or busy. Existing pairings keep working. Try again later rather than pairing again.
* **You paired and got nothing back.** Check you messaged the exact line shown during pairing, and that you have not hit the daily pairing cap.
* **A channel reply looks vague.** That is channel discretion doing its job. Ask the same question in your DM to get the full answer.
# Talk to your Talent with voice notes (/docs/talent/voice-notes)
Sometimes typing the brief is the slow part. You can talk to your Talent instead, and you can ask it to answer out loud.
## Send it a voice note [#send-it-a-voice-note]
Record a voice message and send it to your Talent the way you would to a person, in iMessage, Slack, or Telegram. TalentOS transcribes it and your Talent works from the transcript. You do not have to say anything special first.
This is the fastest way to hand over a messy brief. Walking to a meeting and describing what you need in forty seconds usually beats typing three careful sentences.
Your Talent treats a transcript as information to act on, never as instructions from TalentOS. If a voice note contains something that reads like a command to change its rules, it will not follow it.
## Get a spoken answer back [#get-a-spoken-answer-back]
Ask for it. In a message to your Talent on iMessage, Slack, or Telegram, say:
> Send me that summary as a voice note.
It generates the audio and delivers it as a real voice message in that channel, so it plays inline rather than arriving as a file you have to download.
## The one limit worth knowing [#the-one-limit-worth-knowing]
**One voice note per turn.** If a single run produces several audio clips, you get the first one in the channel and the rest stay in TalentOS, where you can play them in the web console. This is on purpose. A phone filling with audio bubbles is worse than a short honest reply.
## A worked example [#a-worked-example]
You are between meetings. You send an iMessage voice note:
> Read the last message in my inbox from the Acme thread, tell me what they are actually asking for, and read it back to me.
Your Talent reads the thread with your Gmail connection, works out the ask, and replies with a voice note you can listen to in the corridor. The written version is in the same thread if you want to scan it later.
## Playing audio in the web console [#playing-audio-in-the-web-console]
Every clip your Talent generates is saved to your workspace and plays in the web console. If you missed a voice note on your phone, or you got only the first of several, open the conversation in the console and play them there.
## What to check [#what-to-check]
* **You got text saying the audio could not be sent.** The message will say the audio is saved in TalentOS. Open that conversation in the web console and play it there. The written reply is unaffected.
* **You only got one clip.** That is the per turn limit. The rest are in the console.
* **Your voice note was not understood.** Long, noisy, or heavily accented recordings transcribe less reliably. Send a shorter clip, or type the one detail that got lost.
* **Nothing arrives at all.** Check the channel itself is still connected under **Connections**. A voice note cannot arrive on a channel that is disconnected.
# Know who can see your chats and memory (/docs/talent/who-can-see-your-work)
People hesitate to give an AI teammate real work until they know who else is reading it. This is the honest answer.
## Nobody else sees your Talent's work [#nobody-else-sees-your-talents-work]
Your **conversations**, your **personal memory**, your **runs**, your **approvals**, your **connections**, and the suggestions your Talent makes to you are visible to you and to nobody else in your workspace.
That includes admins. It includes your manager. An admin who queries for your conversations gets the same empty result as any other member. This is not a filter someone remembered to apply, it is how the data is stored and indexed, so a new screen cannot accidentally expose it.
## The three narrow exceptions [#the-three-narrow-exceptions]
There are exactly three things about your Talent that reach an admin, and each one deliberately stops short of content.
1. **What a run cost.** An admin can see the credits a run spent, when it happened, whether it succeeded, and which model lane it used. They cannot see what it said, what it read, or what it produced. The workspace pays one bill, so the people who pay it can see what it is made of. A Routine name appears only if that Routine is published to the whole workspace. Anything private shows as "Private routine".
2. **That you have a Talent.** An admin sees your Talent's name, its mascot, and whether it is active on the Team page. Nothing it has ever done.
3. **Bounded support data about nudges.** Whether a proactive check spoke or stayed silent, and when. No message content, no destination, no conversation.
Spend is visible. Content is not. An admin can tell that your Talent had an expensive Tuesday and cannot tell you what you worked on.
## What your Talent itself can reach [#what-your-talent-itself-can-reach]
Your Talent works with:
* **Your own connected accounts.** Not a teammate's.
* **Your personal memory.**
* **Approved company memory**, plus approved memory for your one active department.
A department manager seat is a governance role over shared work. It does not widen anyone's view into a report's private Talent, and it does not widen what a model sees.
## A worked example [#a-worked-example]
Mina asks her Talent to analyze a sensitive renewal and draft options. Her workspace admin can see that Mina used credits and that the run completed, but cannot open the conversation, draft, source material, personal memory, or connection. Mina shares the finished recommendation herself when it is ready.
## In shared Slack channels [#in-shared-slack-channels]
When your Talent replies in a channel other people can read, it treats that audience as real. It will not volunteer your memories, notes, documents, schedules, private conversations, or anything it read from your connected accounts. If the honest answer needs private detail, it keeps the channel reply short and takes the rest to your direct message. See [reaching your Talent from Slack, Telegram, or iMessage](/docs/talent/slack-telegram-imessage).
## Your credentials [#your-credentials]
When you connect an app, the secret behind that connection is never shown again to anyone, including you. You see the connection and its status. You never see the secret itself, and neither does anyone else.
## Sharing on purpose [#sharing-on-purpose]
Nothing private becomes shared by accident. Work reaches your workspace only when you explicitly publish it, for example by promoting something into a Routine and having it published to the Marketplace. Until then it is a private draft.
The **Copy private link** on a conversation is navigation, not sharing. Sending that link to a colleague gets them nothing.
## What to check [#what-to-check]
* **You expected a teammate to see your work and they cannot.** That is correct behavior. Publish it to the Marketplace, or share the output itself.
* **An admin asked you for a transcript.** They cannot pull it. Paste or export what you want to share yourself.
* **Your Talent will not answer something in a channel.** Ask again in your direct message. Channel discretion is doing its job.
* **A connection shows as connected but your Talent says it cannot act.** The connection exists but the action may need your approval, or the account may have lost access on the provider's side. Reconnect it in Connections.
# Run your day from Home (/docs/talent/work-from-home)
Home is your daily workbench. It is built to answer three questions at a glance: what needs you, what your Talent is doing, and what is already scheduled. It is deliberately not an endless chat log.
## The parts of Home [#the-parts-of-home]
* **The composer.** The middle of the screen. This is where you assign work.
* **Today.** A short list of what is live or waiting. It shows a badge only when something genuinely needs you, so an empty Today means you are clear.
* **Upcoming.** Your scheduled Routines, each showing when it next runs.
* **The conversation rail.** On desktop, the left rail holds **Pinned**, **Routine chats**, and **Recent**, with a **Chat history** control at the bottom that opens the full searchable list.
## Task or Chat: pick the right one [#task-or-chat-pick-the-right-one]
The composer gives you two ways to send the same text, and the difference matters.
1. **Task** queues the work right now, in its own conversation. Use it when you want a result.
2. **Chat** opens a conversation with your draft intact and queues nothing. Use it when you want to think out loud first, or when you are not sure the request is right yet.
If you press Task and immediately realize the brief was wrong, say so in the conversation that opens. Your Talent can stop and take the correction.
## A worked example [#a-worked-example]
You want a weekly pipeline note. In the composer, choose **Chat** and type:
> I want a short Monday note on deals that moved in HubSpot last week. Help me decide what should be in it.
You go back and forth for two minutes, agree on the shape, then say "do that now" to run it once. When you are happy with the output, ask your Talent to turn it into a Routine so it runs every Monday morning without you.
## Managing conversations [#managing-conversations]
Every row in the rail opens that exact conversation. On each one you can **pin**, **rename**, or **delete**. A green dot means there is new activity in that conversation since you last looked at it. Pinning only tidies your rail. It never creates a notification.
Once a conversation has been saved, you can use **Copy private link** to bookmark it. That link only opens for you. Sending it to a colleague gets them nothing.
Selecting live work in Today reopens the exact conversation where that work is running. It does not dump you into a global history page.
## Manage, without leaving Home [#manage-without-leaving-home]
The **Manage** panel on Home opens the settings that belong to your Talent rather than to your workspace: **Memory** (what it remembers and where it learned it), **Connections and channels**, **Permissions** (what it may do on its own and what needs your approval), and **Model** (which model lane it runs on). See [choosing a model lane and response style](/docs/talent/model-and-response-style).
## What to check [#what-to-check]
* **A conversation is missing from the rail.** The rail is bounded on purpose. Open **Chat history** and search for it.
* **A conversation link opens Home instead.** The link is malformed, the conversation was deleted, or it belongs to someone else. TalentOS will not tell you which, by design.
* **Today looks empty but you expected work.** Work with no web conversation, such as a Routine that delivers to Slack, will not appear as a live conversation. Look in [Activity](/docs/talent/receipts-and-activity) under Settings for its receipt.
* **Your draft vanished when you opened Manage.** It should not. Manage restores the composer and your text when it closes. If it did not, report it to support with the time it happened.
# Curate department memory (/docs/team-workspace/curate-department-memory)
Department memory is the shared page your department's Talents read before they
work. Get it right once and everyone in the department stops re-explaining the
same thing to their own Talent.
You can curate it if you hold the **manager seat** on that department. Admins
and the workspace owner can curate any active department. Everyone else in the
department proposes, and a curator decides.
## Open it [#open-it]
* **Manager**: sidebar, **Manager** group, **Department memory**. It follows the
department your Manager group is currently on.
* **Admin**: **Team**, **Departments**, open the department, then its memory.
* **Member**: sidebar, **Memory**, then the department scope. They get an add
and edit form that submits proposals.
## The page [#the-page]
Knowledge is filed under six sections, in this order: **Overview and mandate**,
**Priorities**, **People and ownership**, **How we work**, **Tools and
standards**, and the optional **Challenges and dependencies**.
Each entry is a **Fact**, a **Decision**, or an **SOP**. Keep each one to a
single short sentence. A page of clean one-liners beats a wall of prose, because
the Talent quotes the line it needs rather than the whole page.
## Accept a proposal [#accept-a-proposal]
1. Open **Department memory** for the department you manage.
2. Work through the pending proposals. Each one shows the change proposed,
what it targets, the current version of that page, and what differs.
3. Read the delta, not just the new text. If it is nearly right, correct the
wording and the type as you accept it.
4. Accept, or send it back.
Members never see pending proposals and never see who proposed what. They see
the approved page.
## A worked example [#a-worked-example]
Someone in Client Services proposes an SOP: "Weekly client update goes out
Monday 9am from the CRM pipeline, copied to the account lead." You accept it into
**How we work**.
From the next turn, when anyone in Client Services asks their Talent to draft
the weekly update, it already knows the day, the source, and who is copied. You
did not have to touch anybody's Talent to make that true.
## Editing and undoing [#editing-and-undoing]
* Accepting an edit creates a new version. The old one is kept, not overwritten.
* Archiving hides an entry and is reversible. Nothing accepted is destroyed.
* Old links keep working. A link to a superseded entry resolves forward to its
current version.
When a Talent proposes a department fact on its owner's behalf, it always asks
that person for approval first, every time. Shared knowledge is never written
silently.
## What to check [#what-to-check]
**A member sees an empty state.** They hold no department, or their department
was archived. Put them in an active department on the **Directory** tab.
**"This changed since you started."** Someone else edited that entry while your
review was open. Reload the item, read the current version, and decide again.
The conflict is deliberate, so you never accept a change against a page that has
moved on.
**You can read a department but not curate it.** Your manager seat is on a
different department. Seats are per department, and admins get every active one.
**Curation controls are gone entirely.** The department was archived. Archived
departments are closed to writes.
**The whole team still gets the old answer.** Check the entry is accepted rather
than pending, and that the person asking is in that department. Memory is scoped
to the asker's current department, resolved at the moment they ask.
# Give someone a manager seat (/docs/team-workspace/give-someone-a-manager-seat)
A manager seat lets the person who actually runs a department look after what
that department knows, and see how it is doing, without handing them the keys to
billing, invitations, or workspace settings.
Manager is a grant, not a role. It sits on top of whatever the person already
is. Your Ops lead can hold a manager seat as a member. So can an admin who runs
a department day to day.
## Assign a manager [#assign-a-manager]
1. Open **Team** in the sidebar, then the **Departments** tab.
2. Open the department.
3. Choose its manager from the people available.
4. Save.
Each department has exactly one manager. Assigning a new one replaces the
previous manager, so you never have to remove the old one first.
## What the seat adds [#what-the-seat-adds]
A **Manager** group appears in their sidebar, in the daily work cluster rather
than down with the account settings. It holds:
* **Department**, a read view of the department they run: its profile, its
roster, and how the department is tracking.
* **Department memory**, where they curate the shared knowledge their
department's Talents rely on.
If someone manages more than one department, the Manager group works on one
selected department at a time, and it remembers the last one they opened.
A manager seat is scoped to the department it is on. A manager who runs Growth
sees Growth. It grants nothing anywhere else in the workspace, and it does not
change what they can do with their own Talent.
## What a manager still cannot do [#what-a-manager-still-cannot-do]
The seat is department oversight, not administration. Managers do not invite or
remove people, do not move people between departments, do not create or archive
departments, and do not touch billing or workspace settings. Those stay with
admins and the owner.
A manager also cannot see how their reports work. They see the department's
shared artifacts and its state, never anyone's chats, personal memory, or
individual Talent activity. That line is the same for managers, admins, and the
owner.
## A worked example [#a-worked-example]
Priya runs Client Services in a 20 person agency. She is a member, not an admin.
You give her the Client Services manager seat. Now:
* She curates the Client Services page in **Department memory**, so every
Talent in her department answers a client question the same way.
* When one of her team proposes "we always send the CRM pipeline summary on
Monday morning", she reviews it and accepts it, and it applies to the whole
department from that point.
* She still cannot invite the new account manager. She asks an admin for that.
## What to check [#what-to-check]
**The person is not offered as a manager.** They must be an active member of the
workspace. Someone with a pending invitation cannot hold a seat yet.
**The Manager group is missing from their sidebar.** Ask them to reload. The
group appears only for someone holding at least one seat on an active
department, and it disappears the moment the seat goes.
**The manager seat vanished by itself.** Two things clear seats automatically:
archiving the department, and removing the person from the workspace. Both are
intentional. Reassign the seat after you move the department's people around.
**They can see the department but not curate memory.** Confirm the seat is on
the department they are trying to curate, not a neighbouring one. Curation
follows the seat, not their own department.
# Invite someone to your workspace (/docs/team-workspace/invite-someone-to-your-workspace)
By the end of this you will have sent an invitation. When the person accepts,
they get their own Talent, connect their own tools, and start working without
touching anything of yours.
## Send the invite [#send-the-invite]
1. In the sidebar, open **Team** from the group pinned above your profile card.
Team is visible to admins.
2. Go to the **Invitations** tab.
3. Enter one work email address, or paste several at once to invite a whole
department in one go.
4. Choose a role. **Member** is the everyday role and covers the whole product.
**Admin** adds governance: inviting people, departments, workspace settings.
The Admin option appears only if you are the workspace owner.
5. Choose a department. The picker appears for members, who always hold one.
Admins are workspace wide, so they get no picker.
6. Send. Each address goes out as its own invitation and appears in the
**Invitations** tab as pending.
You can also invite straight into one department from that department's detail
page. That form locks the role to member and fills the department in for you.
## What the new person gets [#what-the-new-person-gets]
They sign in, name their Talent, and connect the apps they personally use, for
example their own email and calendar. Their Talent acts as them, with
their access and nobody else's. Connecting email does not put your inbox in
front of them, and it does not put theirs in front of you.
They can browse the **Marketplace** on day one and adopt anything your team has
already published there, which is the fastest way to make a new hire useful in
their first hour.
## Track, resend, revoke [#track-resend-revoke]
The **Invitations** tab lists one row per person, newest first. From a row you
can resend or revoke.
Invitations lapse if they are not accepted. A lapsed invite stays in the list so
you can see it and resend rather than wondering where it went. Resending creates
a fresh invitation and the list keeps showing only the newest one for that email
address.
Revoking an **admin** invitation is owner only. Any admin can revoke a member
invitation.
## Seats [#seats]
Every active person in the workspace is one seat, and admins and members count
exactly the same. Adding someone changes the seat count on the plan, so invite
in the order you actually want to pay for.
## A worked example [#a-worked-example]
An operations lead creates an Operations department, pastes four work email
addresses into one invitation, assigns all four as members, and selects that
department. Three accept the same day and each gets a private Talent. The fourth
stays pending, so the workspace adds three seats, not four.
## What to check [#what-to-check]
**"Create a department before inviting members."** Member invitations require a
department and your workspace has none yet. Create one on the **Departments**
tab, then invite.
**"This department is no longer available."** The department was archived while
your invite form was open. Refresh the page and pick an active department.
**"Only the workspace owner can invite an admin."** Ask the owner to send it, or
invite the person as a member now and have the owner promote them afterwards.
**The invite button is missing or writes are blocked.** Inviting is a write, so
it is disabled while the workspace has no live plan or trial. The owner needs to
sort billing out in **Settings** first, then invites work again.
**Some addresses in a batch went out and one did not.** Each address is sent
separately on purpose, so one typo does not block the rest. Fix that address and
send it on its own.
**The invitation never arrived.** Check the spam folder first, then resend from
the **Invitations** tab. If a person changed email address, revoke the old
invitation and send a new one rather than resending the stale row.
# Organize people into departments (/docs/team-workspace/organize-people-into-departments)
A department is the scope for shared knowledge and for manager oversight. Once
someone is in Client Services, their Talent can draw on what Client Services
knows, and the Client Services manager can curate that knowledge for everyone in
it.
Set departments up before you invite people, because a member invitation needs
one.
## Create a department [#create-a-department]
1. Open **Team** in the sidebar, then the **Departments** tab.
2. Choose **Create department**.
3. Give it a name your team already uses. A 20 person agency usually ends up
with something like Client Services, Growth, and Ops rather than a formal org
chart.
4. Add a short description of what the department does. This becomes the first
thing its shared memory says about it.
5. Save.
Names must be unique. If you get **"A department with this name already
exists."**, an active department already uses that name, including a version
that differs only in capitalization.
## Where people sit [#where-people-sit]
* **Every member holds a department.** They get one when they are invited, and
they always have one after that.
* **Admins can be workspace wide.** An admin does not need a department, though
they can sit in one like anyone else.
* **Only admins move people.** Change someone's department from their row in the
**Directory** tab.
Changing an **admin's** department is owner only. A non-owner admin sees that
control disabled with an explanation, rather than a button that fails.
Moving someone re-scopes their access immediately. If you move a person from
Growth to Ops, their Talent stops reading Growth's shared memory on the next
turn and starts reading Ops. Nothing is copied across, so move people before
they need the new context, not after.
## Read the Directory [#read-the-directory]
The **Directory** tab has two views:
* **People** lists everyone active, admins and members alike, with their role and
department.
* **AI teammates** lists each person's Talent by name, with the department it
works in, the person it belongs to, and whether it is Active, Paused, or still
Finishing setup. A person who has not set theirs up yet shows **No AI teammate
yet**, which is the fastest way to spot a stalled onboarding.
Filter the Directory by a department, by **Managers**, or by **Pending** to find
people who have not finished setting up.
## Archive a department [#archive-a-department]
Archiving retires a department without deleting its history. Open its row on the
**Departments** tab and archive it.
A department can only be archived once it is empty. Move its members somewhere
else and clear its pending invitations first. Archiving also releases the
manager seat on that department.
## A worked example [#a-worked-example]
A 16 person agency creates Client Services, Growth, and Operations. It places
each member in the team they work with, assigns the Client Services lead a
manager seat, and adds "Client updates go out every Monday morning" to that
department's shared memory. When one coordinator moves to Operations, an admin
updates the Directory row before their next Talent turn, so the Talent starts
using Operations context immediately.
## What to check [#what-to-check]
**"Cannot archive, 3 active members and 1 pending invitation."** Exactly what it
says. Move those members to another department on the **Directory** tab, and
revoke the pending invitation on the **Invitations** tab, then archive.
**"Choose a different department name."** A duplicate. Rename the old one or
pick a different name.
**The new department has no manager.** That is normal. Managers are assigned
separately and are optional.
**Someone's Talent still quotes the old department.** Confirm the move landed on
their Directory row. Access is derived from their current membership at read
time, so once the row is right the next turn is right.
# Remove someone from the workspace (/docs/team-workspace/remove-someone-from-the-workspace)
Removing someone ends their access, stops their Talent, and takes their seat off
the bill. Do the two clean-up steps below first, because a couple of things go
away with them.
## Before you remove them [#before-you-remove-them]
1. **Reassign any department they manage.** Removing a person clears every
manager seat they hold, and the department is left without a manager.
2. **Move anything the team relies on into shared ground.** Their private
Routines, Skills, and Artifacts are theirs. Work they published to the
**Marketplace** stays available to the workspace, so publish the important
ones before their last day. A Monday CRM pipeline summary that only lives
in their private Routines stops running when they go.
## Remove them [#remove-them]
1. Open **Team** in the sidebar, then the **Directory** tab.
2. Find the person on the **People** view.
3. Remove them from their row and confirm.
## What happens [#what-happens]
* They lose access to the workspace immediately.
* Their Talent stops, and so does anything only it was running.
* Every manager seat they held is released.
* The seat count on your plan is recalculated straight away.
* Their past spend stays in the workspace usage report, labelled as a removed
member, so the member totals still add up to what you were actually charged.
* Their chats and their personal memory stay private. They are not handed to an
admin, and there is no way to read them after the fact.
## What you cannot remove [#what-you-cannot-remove]
**Yourself.** Removing your own membership is refused. Ask another admin.
**The workspace owner.** "The workspace owner can't be removed." There is exactly
one owner per workspace, and the seat is protected. Contact support if the owner
is leaving the company.
**An admin, unless you are the owner.** "Only the workspace owner can remove an
admin." A non-owner admin can remove members, not admins.
## Demoting instead of removing [#demoting-instead-of-removing]
If someone is staying but should not govern any more, change their role rather
than removing them. Role changes between admin and member are owner only, and
the person keeps their Talent, their work, and their seat.
Two rules apply. You cannot demote the last admin: promote someone else first.
And the owner cannot be demoted at all, including by themselves.
## A worked example [#a-worked-example]
Jordan leaves an agency on Friday. On Thursday, an admin reassigns Jordan's
Client Services manager seat, asks Jordan to publish the weekly pipeline
Routine the team still needs, and then removes Jordan from the Directory. The
seat count drops immediately, the published Routine stays available, and
Jordan's private conversations remain private.
## What to check [#what-to-check]
**"That person is not an active member of this workspace."** They are already
removed, or the directory is showing a stale row. Refresh the Directory.
**The Remove control is missing.** You are looking at an admin's row and you are
not the owner, or you are looking at the owner's row.
**Removal is blocked along with everything else.** Removals are writes, so they
stop while the workspace has no live plan or trial. The owner needs to fix
billing in **Settings** first.
**A department now shows no manager.** Expected. Assign a new one on the
**Departments** tab.