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

# Working Directory

> How the working directory gives the Synti agent direct, local filesystem access to a project folder.

The working directory is the folder the agent treats as its default context — the equivalent of `cwd` in a terminal. Every file read, edit, search, and command the agent runs starts from this location unless you point it somewhere else. Setting a working directory is the fastest way to let the agent work directly with a real project on your machine.

## What the agent can do

Within the working directory, the agent operates on your local files using its built-in tools — no external connection or authentication required:

* **Read, write, and edit files** in place.
* **Find files** with Glob patterns and search their contents with Grep.
* **Run shell commands** in the directory's context.
* **Use git** when the folder is a repository — inspecting history, branches, and diffs.

These are native agent tools, which is what makes filesystem work fast and direct.

## Setting the working directory

### Per conversation

Click the folder badge in the composer to set or change the working directory for the current [conversation](/core-concepts/conversations). The badge shows the folder's name and, if it's a git repository, the active branch. Changes take effect immediately for subsequent commands, and the choice is saved with the conversation.

### Workspace default

To apply the same starting point to every new conversation in a workspace, set a **Default Working Directory** under **Settings → Workspace Settings**. New sessions in that workspace inherit it automatically, and you can still override it per conversation.

## Automatic project context

If the working directory contains a `CLAUDE.md` file, Synti automatically loads its contents into the conversation. This is a convenient place to record project structure, build and test commands, conventions, and team guidelines so the agent follows them without being told each time.

<Tip>
  Keep a short, current `CLAUDE.md` at the root of your project. It's the single most effective way to make the agent's output match how your project actually works.
</Tip>

## Working directory vs. sources

The working directory and [sources](/getting-started/introduction) solve different problems, and they complement each other:

|        | Working directory          | Sources                            |
| ------ | -------------------------- | ---------------------------------- |
| Access | Direct local filesystem    | External services and data         |
| Auth   | None required              | API keys or OAuth                  |
| Tools  | Native agent tools         | MCP / REST tools                   |
| Scope  | One folder on your machine | Remote services and connected data |

Use the working directory for the project you're actively editing, and use sources to reach out to GitHub, issue trackers, or other systems.

## Best practices

* **Point at a specific project folder**, not a broad location like your home directory, so the agent stays focused and safe.
* **Lean on git** when the folder is a repository — the agent can branch, diff, and commit as part of its work.
* **Combine with permission modes.** The working directory decides *where* the agent works; [permission modes](/core-concepts/permissions) decide *what* it's allowed to do there.

<Note>
  Because the working directory grants direct file access, pair it with an appropriate [permission mode](/core-concepts/permissions) — start in Explore to review changes before allowing edits.
</Note>