# Working in Buff Code

> Everyday work in Buff Code: a session with the agent, changes and undo, plan and goal, side conversation, Git, automations, workflows, extensions.

This section is about everyday work: how to give the agent a task, follow what it does,
check and undo changes, run large tasks and hand over repetitive work.
Installation, sign-in and what the organisation provides are described in the neighbouring sections: [Installation](/en/docs/buff-code/install),
[Sign-in and working context](/en/docs/buff-code/sign-in), [Models and usage](/en/docs/buff-code/models).

Buff Code is a **window** (an app for Linux, macOS and Windows) and the `buff` **console**. They share one
core: sessions, pins, automations, workflows, extensions and the language are common. The pages describe
the window; what is done and how in the console is at the end of each page.

_Illustration: The Buff Code window: sessions on the left, the conversation with the agent on the right and, below it, a summary of what it changed._

## Where to start

1. **Open the project folder** — on Home, “Projects” → the folder icon.
2. **Describe the task** in the input field. The agent reads the code, makes edits, runs the tests and
   tells you what it did.
3. **Check the changes** — under the answer there is a “Changed N files” summary; a click opens the diff.
   Don’t like it — “Rewind the turn”.
4. **Commit** — the branch chip under the input field opens the Git window; the agent can
   suggest the commit message.

## Pages in this section

| Page | About |
| --- | --- |
| [The window and the session list](/en/docs/buff-code/work/window) | Home, the session column, pinning, groups, archive, the session state in a word, searching the conversations |
| [Session: input field, modes, questions](/en/docs/buff-code/work/session) | the “+” menu, permission modes, permission requests, the agent’s questions, attachments and quotes, background work, balance |
| [Changes and undo](/en/docs/buff-code/work/changes) | the turn summary, the changes panel, safe undo, the file tree and file view |
| [Plan, to-do list and goal](/en/docs/buff-code/work/plan-goal) | a plan with approval, the agent’s to-do list, the `/goal` goal with a check after every turn |
| [Side conversation](/en/docs/buff-code/work/side) | a question along the way without derailing the main task |
| [Git in the window](/en/docs/buff-code/work/git) | changes, a commit with a message from the agent, push, branches, history |
| [Automations](/en/docs/buff-code/work/automations) | scheduled tasks for the agent — even with the window closed |
| [Workflows](/en/docs/buff-code/work/workflows) | subagents working in phases with a result; save and run again |
| [Extensions](/en/docs/buff-code/work/extensions) | MCP servers, skills, commands, subagents, memory, hooks, trust |
| [Settings and first launch](/en/docs/buff-code/work/settings) | introduction, language, theme, text size, keys, proxy |

## Everything is recorded

Every turn — yours, an automation’s or a workflow’s — is recorded in the organisation’s data just like
an ordinary development session: it is visible in the workspace, you can ask questions about it, and usage is charged to
the organisation’s plan. In detail — [Development sessions](/en/docs/buff-code/sessions).