AGENTCONN
A

// Agent profile

Atlas

Coding Free

About Atlas

pacifio/atlas is source control built for the multi-agent coding era. Every agent run produces checkpoints: commits linked back to the session that made them alongside prompts, tool calls, and reasoning. Run Claude Code, Codex, or any agent from the ACP registry side by side against the same codebase with shared memory, so switching agents mid-task doesn't mean starting over. Atlas hit 3,000+ stars on GitHub and has been trending consistently since its May 2026 launch. Written in Rust, it runs local by default — code, notes, and sessions stay on your machine — with optional team sync via organizations. The shared memory layer means a decision Claude Code made shows up in Codex's next prompt, eliminating the cold-start problem that plagues multi-agent workflows.

Key Features

  • Checkpoint-based source control — every commit links back to the agent session, prompts, tool calls, and reasoning that produced it
  • Multi-agent side-by-side — run Claude Code, Codex, and ACP registry agents against the same codebase in one window
  • Shared agent memory — decisions, plans, file changes, and architecture notes flow between agents automatically
  • Semantic @ mentions — resolve files, folders, symbols, branches, commits, notes, and past sessions locally before the prompt is sent
  • Session handoff — curated fact pack plus tail of your last session feeds into the next one, even across different agents
  • Local by default — all data stays on your machine, with optional org-level sync for teams

Overview

Atlas is source control for coding agents from pacifio, currently at 3,000+ GitHub stars and climbing fast. The core insight is that agents now write a large share of code but keep none of the reasoning behind it. Atlas records both the changes and the reasoning, then makes them queryable: what changed, by whose agent, at what point in time, and why.

Key Capabilities

The differentiator is the checkpoint model. Every agent run produces a checkpoint that links a commit back to the full session — prompts, tool calls, file diffs, and reasoning. This is not just git history with extra metadata; it is a structured record that lets you trace any line of code back through the agent conversation that wrote it.

The second differentiator is shared memory across agents. Atlas maintains a persistent on-device memory index of decisions, plans, file changes, and architecture notes. When you switch from Claude Code to Codex mid-task, the new agent picks up context from the last one automatically. The semantic matching runs locally — your message is embedded on-device and matched against the project’s memory index every turn.

Atlas supports the Agent Client Protocol (ACP) from Zed, so any compatible agent can plug in. Claude Code and Codex run as external subprocesses over ACP. Atlas’s own built-in agent runs in-process on Cersei, their Rust agent framework.

Use Cases

  • Teams running multiple coding agents and needing a single audit trail of what each agent changed and why
  • Solo developers switching between Claude Code and Codex who want context to carry across sessions without manual handoff
  • Engineering leads reviewing agent-generated PRs who need to trace code back to the reasoning that produced it
  • Organizations building agent workflows that require persistent memory across sessions and agent types

Similar Agents