---
title: Start a session
description: Every control in the Build composer — repository, branch, mode, quality, budget, attachments and review options — and how to write a brief Ysra can hold itself to.
---

Open **New session** (++cmd+n++ / ++ctrl+n++) or the **Build** tab on the home
screen. One brief starts the work.

![The Build composer with an empty brief field, the repository and branch chip, a Branch button, the quality selector set to Beta, the budget and the send button](../assets/screens/home-composer.webp){ width="1600" height="1280" }
/// caption
The composer. The line under the brief summarises what the session will use:
repository, mode, quality, reasoning, compute, and the branch it will write to.
///

## The controls

| Control | What it does |
|---|---|
| **Brief** | What should be true when the work is done. Paste a long brief freely — it becomes a chip you can expand. |
| **+ Attach** | Add files and images from your workspace or your device: specs, screenshots of a bug, a design to match. Images are analysed and bound to this session. |
| **Repository** | The repository to work in. Only repositories you granted to Ysra appear. See [GitHub](../integrations/github.md). |
| **Branch** | The branch to start from. Ysra always writes to a **new branch**; it never edits your base branch. |
| **Quality** | How thorough the session is. Higher quality spends more on review and proof. |
| **Budget** | The ceiling for this session. It shows what your wallet has available; Ysra stops before exceeding it and asks. |
| **Agent / Plan first** | *Agent* starts immediately. *Plan first* produces a specification to review and approve before any code runs (where available on your account). |

Advanced options — in the session launcher and on the portal's New session
dialog — let you choose:

- **Compute profile** — *Balanced* for everyday work; a more intensive profile
  for large or hard changes.
- **Reasoning depth** — how much deliberation each step gets.
- **Acceptance review** and **browser review** — whether criteria are reviewed
  independently, and whether interactive behaviour is proven in a real browser.
- **Hosted or local runtime** — hosted by default; local where enabled.
- **Delivery** — what should happen to the candidate when it's done, within your
  [autonomy policy](../workspace/autonomy-policy.md).

!!! tip "Turn on browser review for anything a person clicks"
    Without it, interactive requirements — forms, filters, modals — are reported
    as *unproven*, not passed.

## Start from a playbook

The **Start from** cards under the composer are [playbooks](playbooks.md):
repeatable briefs for common jobs — a weekly dependency bump, production error
triage, a two-phase schema change, a flaky-test quarantine. Choosing one fills
the composer; you can edit before sending.

## Write a brief Ysra can hold itself to

Ysra turns your brief into **acceptance criteria** — a checklist it is reviewed
against at the end. The clearer the brief, the sharper the checklist.

**Describe outcomes, not steps.**

```text title="Weak"
Add pagination.
```

```text title="Strong"
Paginate the invoices page: 25 per page, archived invoices sorted last.
Keep the URL shareable (?page=3). Existing filters must keep working.
Add tests for the first page, the last page, and an empty result.
```

**Name what must not change.** "Keep the public API unchanged", "don't touch the
billing module", "no backend". Exclusions become constraints Ysra is reviewed
against — never extra work.

**Paste the criteria you already have.** If a ticket lists acceptance criteria,
paste them. Your own criteria are reviewed one by one; Ysra only derives
criteria when you didn't give any.

**Be explicit about technology when it matters.** "Use the existing `DateRange`
type" or "must be a Next.js app" is binding. A technology you merely mention in
passing is not treated as a requirement, and "A or B" stays a choice.

**Attach what a person would need.** A screenshot of the bug, the design, the
error log, the API contract.

## What happens when you press send

1. The session opens immediately in the [workroom](workroom.md) with your
   message.
2. Ysra restates the goal and the plan in the conversation.
3. Your budget is reserved and your autonomy policy is frozen for this session —
   changing settings later won't change a running session.
4. Work starts as soon as an isolated workspace is ready.

## Budgets

The budget is a hard ceiling, not an estimate:

- Ysra stops **before** the next paid step that would exceed it and asks you.
- From the portal or the mobile app you can extend it, steer instead, or stop.
- Unused budget returns to your wallet.

See [Wallet and usage](../workspace/wallet.md).

<div class="phones" markdown>

![The mobile budget gate: "Spend $2 more to finish this task?" with the task, repository, branch, plan progress, spend, wallet and waiting time, quick top-up amounts and "Slide to add $5.00", "Steer instead" and "Stop here"](../assets/screens/mobile/gate-budget.webp){ width="585" height="1266" }
/// caption
When a session reaches its budget, you decide — here on the phone.
///

</div>