Deja Vu

Deja Vu (the `deja` CLI) is an open-source memory layer for coding agents, published on GitHub, that indexes the session history your agents already save to disk and gives any agent instant recall of it. The project has gathered around 1,300 GitHub stars and is actively developed in 2026.

Overview

Claude Code, Codex, Cursor and 40+ other coding harnesses all store their sessions locally — and forget them. Deja Vu indexes that history (the questions asked, decisions made, files touched and commands run, with credentials stripped during indexing) and exposes it through a terminal search UI, an MCP server and session-start hooks, so a fix you found in one agent resurfaces in another, even months later and even after context compaction. It runs as a single Go binary with no model and no server: search is local and private, optional semantic recall can be enabled with your own embedding endpoint, and sync between machines goes over SSH with no cloud in between. The project publishes its recall benchmarks (97.2% R@5 on LongMemEval-S) and the corpora behind them, and installs in seconds via Homebrew, npm, Scoop, winget or a shell script.

Key Features

• Indexes session history from 42 coding agents into one memory
• Instant local search TUI with filters by agent, project and time
• MCP server and hooks for automatic recall at session start
• Cross-agent handoff — continue a session in a different tool
• Credentials are stripped from the index as it is built
• Single Go binary; no LLM, no server, works offline

💳 Pricing

Deja Vu is free and open source under the MIT license. Optional semantic recall uses an embedding endpoint you configure and pay for yourself.

Who Is It For

Developers who live in AI coding tools and are tired of re-debugging problems they already solved in an earlier session.