> ## Documentation Index
> Fetch the complete documentation index at: https://docs.goosybear.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Assistant

> Tune how Goosy talks to you, when he checks in, how many drafts he writes, and the standing instructions each agent carries.

**Settings → Assistant** is where you tune Goosy himself — how he talks, when he
speaks up, how much he produces per request, and the standing background each
of his agents carries into every run.

<Note>
  These settings apply to everyone on this account, not just you. Changing them
  takes an owner or admin; anyone else gets the screen read-only, with a notice
  at the top saying why the controls will not move.
</Note>

## Goosy

| Setting                      | What it does                                                                                                                                                                             |
| ---------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Goosy's tone**             | How Goosy talks to your team. **Standard** is Goosy as designed — dry, warm, capable. **Professional** is straight down the line, humor off. **Extra dry** is more deadpan, fewer words. |
| **Goosy speaks**             | Reads replies aloud — the per-message "read it to me" action, and voice mode.                                                                                                            |
| **Auto-speak in voice mode** | Replies speak themselves while you are in voice mode, hands-free. It only does anything while **Goosy speaks** is on.                                                                    |

## Working style

| Setting                                     | What it does                                                                                                                                                                                |
| ------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Voice-first capture**                     | Interviews and check-ins open in talking mode. Typed is always one click away.                                                                                                              |
| **Proactive check-ins**                     | Goosy opens with stalled drafts, upcoming slots and what is working. **Off**, **Daily**, or a **Weekly digest**.                                                                            |
| **Quiet hours**                             | A start and end hour that check-ins are held through. Each person's window is read in **their own** time zone, so two people in different cities can pick the same hours and both be right. |
| **Auto-close Goosy after applying an edit** | The panel closes on its own when an edit finishes and your message box is empty; a started reply keeps it open. **This device only** — it is the one setting here that is not account-wide. |

<Warning>
  Check-ins are not running yet. Your cadence and quiet hours save now and take
  effect when check-ins ship. The screen tells you the same thing under those
  two controls.
</Warning>

## AI behavior

| Setting                | What it does                                                                                                                 |
| ---------------------- | ---------------------------------------------------------------------------------------------------------------------------- |
| **Generation quality** | **Fast**, **Balanced** or **Best**. Best takes longer and costs more credits per piece.                                      |
| **Drafts per request** | How many variants Goosy writes before showing you the best — one, two or three.                                              |
| **Quality scoring**    | Shows each draft's score against your calibration wherever it appears. It changes what you see, not whether scoring happens. |

<Warning>
  Generation and scoring do not read these three yet. Your choices save now and
  take effect when they do. The screen tells you that in place as well.
</Warning>

## Agent instructions

Below the preferences, one panel per agent. Each holds standing background that
agent carries into every run: it is **added to what the agent already knows,
never a replacement**, and it applies from the next run without restarting
anything.

Each agent shows **Default** until you write something, then **Instructed**.
Leave the box empty for none; up to 4,000 characters each. Write the standing
rule rather than the one-off — "never promise outcomes we can't evidence"
belongs here; "make this one shorter" belongs in the request.

Owners and admins write these. There is no way to hand the permission to anyone
else today — **Settings → Account** offers only the automations and connections
grants — so everybody else reads the boxes without being able to change them.

## Research

Two links out to the research screen, where those settings actually live:

* **Research watchlist** — the channels every research run studies before it
  ranks anything. Add one, pause one, or drop one it should stop reading.
* **Research engine** — how much staying on-brand counts when topics are
  ranked, and what a cycle may spend reading your space.

## Where the numbers live

Nothing on this screen reports what you have spent. Credits and the three-month
trend are on **Settings → Usage**, which is the one place that story is told.
See [Usage](/account/usage).