atlas: Source control that links AI agent sessions with commits to record the 'why'

Track code written by AI coding agents alongside the reasoning behind it. Each commit is automatically linked to the generating session, prompts, and tool call logs, saved as a 'checkpoint'. Even months later, you can instantly check and converse about which agent made a change and why.

Run multiple tools in parallel on a single codebase, including Claude Code, Codex, and the Atlas native agent. On-device shared memory persists across agent switches, so decisions or failure records from one agent are automatically reflected in the next prompt for another. Work context is not reset even if sessions are interrupted or tools are switched.

All data processing and embedding search are performed on the local device. Code, notes, and session logs are not sent externally; create an account only when team synchronization is needed. Existing CLAUDE.md or AGENTS.md files and Markdown notes in the .atlas/knowledge/ directory serve as a unified index, designed to allow agents to naturally reference project knowledge.