deja-vu

The most frustrating flaw of coding agents is not that they write bad code — it is that they keep re-debugging a bug you fixed in March. Most memory tools start from a blank page and record forward, but the history that actually matters already sits on disk in the session files that Claude Code, Codex, Cursor, and two dozen other agents have written. deja-vu does one thing: it indexes those existing files into a single shared memory layer every agent on the machine can read.

Its standout capability is recall at the point of action. Before an agent edits a file, runs a command, or hits a failure, deja-vu slides the relevant prior decision, the last working invocation, or the post-error fix it recorded before — automatically, with no one to ask. What it indexes goes beyond conversation text: the files each turn opened, real command exit codes, and the exact spans an edit replaced. Those are the details every context compaction discards first. On the privacy side, keys, tokens, JWTs, and private-key blocks are stripped at index time, and everything runs locally.

If you switch between multiple coding agents daily, maintain a project that spans weeks or months, or are tired of the loop of re-solwing a problem you already fixed, deja-vu is worth a look. It does not replace your agent; it gives every agent on the machine a shared, searchable, action-triggered long-term memory.