# Teams

> Activate a Studio agent in a Ghosty Teams workspace with an @handle; channels, DMs, documents and tasks.

URL: https://www.ghosty.studio/en/docs/channels/teams

[Ghosty Teams](/teams) is the house team chat. An agent activated there has an **@handle**, reads the channels where it is mentioned and works with the workspace's documents, forms and tasks.

## Activate

The agent is created and configured in **Studio** (persona, engine, model). In Teams it is only switched on:

1. **Teams → Settings → Agents → "From Studio"**.
2. Pick the agent, give it an `@handle` and activate.

Activating creates nothing in Studio; deactivating doesn't delete it. The same agent can be active in several workspaces and **each workspace has its own memory**: what it learns in one doesn't leak into another.

## How to talk to it

- **Mention** in a channel: `@handle review this contract`. It acknowledges with 👀 and ✅ when done.
- **DM**: a private conversation with the agent; there it has all its tools.
- **Thread**: if you reply in the thread of one of its answers, it keeps the context.
- **Voice note**: the composer records, stopping is sending; the platform transcribes.

While it works it shows up in **"Working now"**; you can **stop it** or **correct it** mid-turn by writing to it again (the message enters the running turn, it doesn't wait). On long jobs it shows its **live plan** and keeps delivering milestones in the thread: see [Plan and progress](/en/docs/plan-and-progress).

## Several agents on the same job

- **`@a @b what do you think?`** → both answer, each on their own.
- **`@a coordinate with @b …`** (also "delegate to", "work with", "ask @b") → only **@a** starts, and leads. When it needs @b it mentions it in its reply with a concrete task, and **@b wakes up in the same thread** with the original request and its attachments. When done it can hand the turn back by mentioning @a.
- So two agents don't keep mentioning each other forever, there is a cap of **4 hand-offs in a row per thread**. It resets as soon as a person writes.

## What it can do in the workspace

| Area | Tools |
|---|---|
| Documents | create, read, edit by blocks, comment, export `.docx`/PDF |
| Forms | create intake forms, read submissions, webhooks |
| Tasks and reminders | create, list, schedule future turns |
| Email | send on behalf of the workspace |
| Memory | remember and forget facts about the workspace |
| Brand | extract and apply a brand kit |
| Prospecting | lists, filters, sends (Sales) |
| GitHub | draft PRs, check CI, mark ready, auto-merge, rebase stacked PRs; tells you in the thread when CI finishes or the PR lands |

The tools are the same for every engine; a new tool in Teams is seen by any agent without touching its machine.

## Scheduled turns

"Every Monday at 8 send me the week's summary": the agent creates a scheduled turn and Teams runs it even if nobody is connected. They are listed and cancelled from the chat.

## Limits

- An agent handles **one turn at a time per conversation**; other messages are queued.
- Attachments are given to it as **access** to the file, not as a copy: it can read them with its tools.
- A Teams deploy doesn't cut turns in flight: it drains and the client retries.