# Atlas: Source Control for AI Coding Agents

This repository profile is provided by osrepos.com, an open source repository discovery platform.

Source: osrepos.com
Repository profile: https://osrepos.com/repo/pacifio-atlas
Generated for open source discovery and AI-assisted research.

Atlas is an innovative source control system designed specifically for AI coding agents. It allows developers to manage multiple agents, track their changes, and query their activities in a unified environment. Built in Rust, Atlas ensures that every agent's commit is linked to its session, providing full context including prompts and tool calls.

GitHub: https://github.com/pacifio/atlas
OSRepos URL: https://osrepos.com/repo/pacifio-atlas

## Summary

Atlas is an innovative source control system designed specifically for AI coding agents. It allows developers to manage multiple agents, track their changes, and query their activities in a unified environment. Built in Rust, Atlas ensures that every agent's commit is linked to its session, providing full context including prompts and tool calls.

## Topics

- ai
- coding-agents
- git
- version-control
- rust
- developer-tools
- self-hosted
- ai-coding-assistant

## Repository Information

Last analyzed by OSRepos: Thu Sep 17 2026 00:54:13 GMT+0100 (Western European Summer Time)
Detail views: 0
GitHub clicks: 1

## Safety Notice

OSRepos shares public repositories for knowledge and discovery only. Review source code, dependencies, licenses, and security implications before running or installing anything.

## Content

## Introduction

Atlas is an innovative source control system meticulously engineered for AI coding agents. With 4807 stars and 289 forks, this Rust-based project by pacifio addresses a critical gap in modern development workflows: the lack of comprehensive tracking and context for code generated by AI. It enables developers to seamlessly integrate and manage multiple coding agents, track their changes, and query their activities from a single, unified interface. Atlas ensures that every commit made by an agent is deeply linked to its original session, preserving crucial context like prompts, tool calls, and the underlying reasoning.

## Why use and benefits

The rise of AI in code generation has introduced new challenges, primarily the loss of context and reasoning behind agent-produced code. Traditional editors and version control systems fall short in addressing these issues, leading to agents starting from scratch in every session, losing valuable context when switching between different agents, and scattering critical information across multiple locations.

Atlas tackles these problems head-on with a set of powerful features:

*   **Every Commit, Explained (Checkpoints):** Unlike standard Git commits, Atlas creates "checkpoints" that link each commit back to the specific agent session that produced it. This includes the prompts given, tool calls made, and the agent's reasoning, all preserved and queryable long after the code is written. This means you can understand not just *what* changed, but *why*.
*   **Run Any Agent, Side by Side:** Atlas supports a wide array of agents, including Claude Code, Codex, its own native agent, and others from the ACP registry. You can run these agents concurrently against the same codebase, with shared memory, eliminating the need to restart tasks when switching agents.
*   **One Memory, Every Agent:** Agents within Atlas share a unified memory. Decisions made by one agent, such as Claude Code, are automatically accessible to another, like Codex. This shared context, including plans, file changes, failures, and architectural notes, is matched on-device against your current queries.
*   **Your Notes Are Agent Context:** Markdown files in `.atlas/knowledge/`, along with existing `CLAUDE.md` and `AGENTS.md` files, are automatically fed into every agent in the project, enriching their understanding and context.
*   **Smart Context Injection:** Use `@` mentions to reference files, folders, symbols, branches, commits, notes, papers, and past sessions directly into a prompt. These references are resolved locally before the prompt is sent, ensuring relevant context without overwhelming the agent.
*   **Local by Default:** Prioritizing privacy and control, Atlas keeps all code, notes, and sessions on your machine. Account sign-in and organization creation are optional, used only for syncing across teams and devices.

Beyond agent-specific features, Atlas provides a comprehensive workspace with capabilities like a CodeMirror editor, integrated Git functionality, a block terminal, a knowledge base, research tools, a native WebKit browser, and spatial boards for notes.

## Installation

To get started with Atlas, you can download the latest build directly from the official website or the GitHub releases page.

*   **Download:**
    *   Visit the [Atlas Website](https://www.tryatlas.cc/){:target="_blank"} (macOS .dmg, Windows .msi)
    *   Or check the [GitHub Releases](https://github.com/pacifio/atlas/releases){:target="_blank"} for specific versions.
*   **Supported Platforms:** macOS 13+ and Windows 10+ (x64). Linux builds are available but currently untested.

### Build from source

For those who prefer to build from source, Atlas requires **Bun** and **Rust** (stable, via rustup), along with **Xcode Command Line Tools** on macOS or **MSVC build tools** on Windows. Linux users will also need specific GTK 3, WebKit2GTK 4.1, and GLib headers.

**Linux System Dependencies (Example for Debian/Ubuntu):**
bash
sudo apt install -y libglib2.0-dev libgtk-3-dev libwebkit2gtk-4.1-dev


**General Build Steps:**
bash
git clone https://github.com/pacifio/atlas
cd atlas
bun install
bun run dev:app

The initial Rust compilation may take a few minutes, subsequent builds are much faster.

## Examples

Atlas enhances how agents operate by running them as they are and enriching their environment. It supports agents like Claude Code and Codex via the Agent Client Protocol (ACP) as external subprocesses, while its native agent runs in-process. Any agent from the ACP registry can be spawned, with Atlas automatically pulling in their official binaries.

Before your message reaches an agent, Atlas intelligently assembles context around it:

*   **`@` mentions:** Locally resolved in Rust, inlining notes, skills, papers, and past sessions, or resolving files and folders to a path.
*   **Shared agent memory:** Includes active plans, decisions, file changes, failures, and architectural notes, written by any agent.
*   **Semantic matches:** Your message is embedded on-device and matched against the project's memory index.
*   **Session handoff:** A curated fact pack plus the tail of your last session in this project, even if from a different agent.
*   **Existing knowledge:** Knowledge notes, `CLAUDE.md`, `AGENTS.md`, Claude Code's memory files, and Codex's history, all folded into one index.

This approach ensures:
*   **Unified workflow:** Run your existing Claude Code or Codex subscription through Atlas for enhanced context without changing your workflow.
*   **Cross-agent visibility:** Claude Code's memory is visible to Codex, and vice versa, a capability not present in standalone agents.
*   **Efficient context handling:** Referencing large files or folders with `@` sends a pointer, allowing the agent to read on demand, conserving context window space.
*   **Local privacy:** All embedding and retrieval operations occur on your machine, ensuring data privacy.

## Links

*   **GitHub Repository:** [https://github.com/pacifio/atlas](https://github.com/pacifio/atlas){:target="_blank"}
*   **Official Website:** [https://www.tryatlas.cc/](https://www.tryatlas.cc/){:target="_blank"}
*   **Documentation:** [https://docs.tryatlas.cc/](https://docs.tryatlas.cc/){:target="_blank"}
*   **Discord Community:** [https://discord.gg/GmnFggaPfP](https://discord.gg/GmnFggaPfP){:target="_blank"}
*   **License:** [Apache License 2.0](https://github.com/pacifio/atlas/blob/main/LICENSE){:target="_blank"}