deja gives six AI coding tools one searchable work history
deja was built to stop Codex or Claude Code from solving the same problem again after it had already been fixed on another computer. In this case, 3.3 GB of past conversations were stored locally but could not be searched together or seen by the AI tools. deja is one executable written in Go that turns those existing records into a searchable index, without contacting an AI model, using a cloud service, or requiring a .
The first indexing run takes about 10 seconds, while later searches take 7–9 milliseconds. It can find a phrase across every past session, and its MCP feature lets an AI check earlier work when a problem looks familiar. It also adds recent project information when a new session begins and removes , , and JWTs while building the index.
Records can move between computers through SSH, and an old session can be reopened in the tool where it originally ran. It combines histories from Claude Code, Codex CLI, opencode, Cursor, Gemini CLI, and aider; similar tools such as cass, ctx, and claude-mem already exist.
Key points
- It combined 3.3 GB of previously separate histories into one searchable index.
- It supports Claude Code, Codex CLI, opencode, Cursor, Gemini CLI, and aider.
- Initial indexing takes about 10 seconds; later searches take 7–9 milliseconds.
- An AI can use MCP to recall earlier solutions, and new sessions can receive recent project information automatically.
- Secrets are removed during indexing, and SSH can transfer the history between computers.