---
title: "Local setup · Docs · SourceAnt"
description: "Documentation: Explore code and knowledge, manage skills, and review local changes with SourceAnt."
url: "https://sourceant.ai/docs/local-setup/"
---

# Local setup

Use SourceAnt locally to explore your code and knowledge, manage skills, and check changes against them before sharing your work.

To run the review API on your own server, see [Self-hosting](https://sourceant.ai/docs/self-hosting).

## Install

One command, on Linux and macOS:

```
curl -fsSL https://raw.githubusercontent.com/sourceant/cli/main/scripts/install.sh | sh
```

Pin the CLI release with `SOURCEANT_VERSION`. To read the script before running it, open the same URL in a browser first.

## Set up and start

1.  **`sourceant setup`**: installs the agent and a core, writes your config, and creates the index directory. Choose the runtime with `--runtime docker` or `--runtime python`.
2.  **`sourceant ui`**: starts the agent if it is not already answering and opens the panel in your browser. Pass `--no-open` to start it without the browser.
3.  **Add a repository** from the panel, then let it index.

The CLI, agent and core have independent releases. Pin the agent and core with `sourceant setup --core-version 1.0.0-beta.2 --agent-version 1.0.0-beta.2`, replacing the example versions with the releases you want. For a custom runtime, use `--image` with Docker or `--from` with Python; neither can be combined with `--core-version`.

## Commands

-   `sourceant status`: agent version, whether the core answers, restart count, last exit.
-   `sourceant repos`: the repositories currently indexed.
-   `sourceant graph <repository>`: node and link counts by kind and edge type.
-   `sourceant architecture <repository>`: indexed components and dependencies. See [Architecture inspection](https://sourceant.ai/docs/architecture#cli) for depth, test inclusion and baseline comparisons.
-   `sourceant stop`: stop the agent and its Python core or Docker container, keeping the index and configuration.
-   `sourceant version`: build, commit and time.

The agent listens on `http://127.0.0.1:8930`. Override it with `SOURCEANT_AGENT_URL`, or per command with `--agent`. Add `--json` to any command for machine-readable output.

## Connect your coding tools

Point any MCP client at `http://127.0.0.1:8930/mcp`. That gives your assistant the code, knowledge, topology and requirement tools, plus the `review`, `context` and `remember` prompts. Because this instance reaches your checkout, it can also review work you have not committed yet.

Models are called from your machine, so you configure your own provider keys and nothing is billed by us.

Stored skills are also available as MCP prompts. See [Use a skill in a coding tool](https://sourceant.ai/docs/skills#use) for selecting one by name or task.