vshulcz/deja-vu

The core idea here is straightforward and genuinely useful: your AI coding agents produce session logs full of solved problems, and right now those logs are write-only. deja-vu indexes them and gives you semantic recall across sessions and across agents via an MCP interface. That means when your next agent session hits a wall, it can pull the answer from a previous run without you having to remember which session had it or re-run the whole chain. The cross-agent angle is the differentiator. Most session-history tools are single-agent and single-session. This one treats the entire history of your local agent work as a searchable corpus. Zero-dependency binary and SSH portability mean it fits into a remote dev box setup without ceremony. Reservation: the project is early and the quality of recall will depend heavily on how verbose and structured your agents' logs actually are. Sparse logs mean sparse results. -> Best for: AI engineer or solo founder running multiple coding-agent sessions and tired of solving the same problem twice.