"deja", cross-agent memory recall over your own coding agents' session logs, 419 stars in 6 days

If you run Claude Code, Codex, or any of the ten other supported coding agents, they already write session logs to your machine. JSONL files, SQLite databases, sitting there. You solved a problem three weeks ago, your agent figured out a tricky API call last month, and none of it is searchable. Until now.

deja indexes those logs and lets you ask, “Did my agent already solve this?” It runs locally, redacts credentials automatically, and works through a CLI or the Model Context Protocol so your current agent can search its own history. No cloud upload, no new workflow. It reads what already exists.

The project picked up 419 stars in six days, which sounds like hype until you check the commit log. Twenty releases in under a week. That pace usually means someone is fixing real bugs from real users, not farming attention. The tool is rough, the docs are thin, but the core idea is sound. If you are paying an agent to solve the same problem twice because you forgot you already asked it in a different session, you are wasting money and context window.

This is the kind of memory layer that belongs in an AI command centre, the sort of thing we wire into Omni when a team wants agents that learn across sessions instead of starting cold every time. Right now deja is a developer tool, but the pattern applies anywhere you have agents doing repeat work. Index the logs, make them searchable, stop solving the same thing twice. It won’t make headlines, but it will save you hours.