One tool reconstructs shared memory across 33 coding agents

deja-vu builds one shared memory across 33 coding agents, and it does it without asking anyone to change how they work. Claude Code, Codex, Cursor, Copilot CLI, OpenClaw and the rest already write session logs as a side effect of running. deja-vu reads those logs and reconstructs a memory the tools can share, instead of shipping a new store you'd have to feed by hand.

That matters most for anyone who switches tools mid-project. Right now, context built up in a Claude Code session doesn't carry into Cursor or Codex unless you copy it over yourself. deja-vu's bet is that plugging into 33 tools at once beats asking any one team to change its logging format.

Cross-vendor memory tools have a track record problem. Prior projects have died asking developers to adopt a new format before delivering anything back. deja-vu skips that step: if a tool already logs its sessions, which most of them do, there's nothing to opt into.

The project is MIT licensed and has picked up 878 stars since going up on July 14, which is real interest but not proof the reconstruction holds up three months and a dozen tool-switches into a real project.

The open question is fidelity. Session logs weren't designed to be a memory store. They're incidental output, and whatever deja-vu extracts from them is only as good as what each tool happened to write down. Whether it holds context across a tool that logs sparsely versus one that logs everything will decide if this works for a given stack or just the tools it was built against first.