Deja-vu is a zero-dependency, local-first binary that indexes and makes searchable the historical sessions of various AI coding agents. It acts as a universal memory layer, allowing agents to recall past solutions and context efficiently, even from before its installation.

deja-vu

Active·★ 1.3k·MIT·Updated 2026-10-27

deja (deja-vu) is a zero-dependency binary that turns the conversation histories your coding agents already write to disk into a searchable memory layer. The README lists Claude Code, Codex, opencode, aider, Gemini CLI, Cursor, Antigravity, Grok Build, Qwen Code, pi and Copilot CLI as sources. Because it indexes existing history, it starts full instead of recording forward: months of past sessions are searchable the moment you install it. The README claims about 12 ms search over 3.3 GB, which is the project's own figure.

Features listed: `deja "query"` search from the command line; an MCP `recall` tool so an agent can find that a problem was solved weeks ago, even in a different agent; `deja sync ssh <host>` to move memory between machines with no cloud in between; `deja handoff --to codex` to package live context for another agent; a SessionStart hook for automatic recall; `deja share` for a sanitized digest of a session; `deja blame` to see which sessions touched a file; `remember` for durable decisions; and optional semantic search via a local Ollama or LM Studio.

On privacy, the README says API keys, JWTs and private keys are stripped at index time, `deja forget` removes sessions and records tombstones so they are not re-indexed, project exclusions are supported, and nothing leaves your machine unless you sync or share it. Install with a shell script, `go install`, `npx @vshulcz/deja-vu` or Homebrew, then `deja install --all` to wire the MCP server into agents it finds (it keeps a `.bak` of edited configs). It is MIT licensed and was last pushed on 2026-10-12. This review is based on the README and repository data; we have not run it ourselves.

#AI Agent Memory#Session Search#Code Assistant#Local-first#Knowledge Retrieval#Developer Tooling#Context Management#Sensitive Data Redaction

Features

01Fast, Retroactive Session Search

02Automatic AI Agent Memory Recall

03Cross-Machine Memory Synchronization

04Seamless Agent Context Handoff

05Automated Sensitive Data Redaction

Why choose it

+Indexes the session histories of a dozen agents (Claude Code, Codex, Cursor, Gemini CLI and more) that are already on your disk, so past work is searchable immediately

+MCP `recall` tool lets one agent find what another solved, plus `handoff` to continue a task in a different agent

+Local and zero-dependency; secrets are redacted at index time, and `forget` and project exclusions give control over what is kept

+Sync between machines goes over SSH with no cloud in the middle; MIT; last push 2026-10-12

Trade-offs

!It reads all your agent transcripts, which can contain proprietary code and sensitive discussion; redaction is pattern-based, so review `exclude` rules and what you share

!`deja install` edits agent config and guidance files (with `.bak` backups); use `--no-guidance` to opt out of the guidance part

!The speed figures are the project's own, and some agents' indexing needs the `sqlite3` CLI

Maintenance check

ActiveChecked: 2026-10-23

Last commit: 2026-10-23
Commits, last 90 days: 100+
Releases, last 90 days: 30
Latest release: nightly (2026-10-23)
Open issues: 27
Recent issues with at least one comment: 4/18

Commits per week, last 12 weeks

Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.

Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.

Most discussed open issues

#534 Provenance for agent-written code: which prompt produced this commit4 comments · 2026-07-29
#5318 Gemini --resume writes each tool result again, and deja shows and indexes it twice3 comments · 2026-10-18
#5588 cline uninstall leaves the reader's own SKILL.md as SKILL.md.bak instead of restoring it2 comments · 2026-10-22
#5295 deja aider re-reads the whole chat history on every message when two aiders start together: 136 ms per message on a 46 MB history2 comments · 2026-10-18
#4976 Semantic tier: key vectors by content so they survive index rewrites, and let vectors retrieve on their own2 comments · 2026-10-12

Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.

Compatibility

MacOS

Verified via docs

Linux

Verified via docs

Windows

Verified via docs

Runtime

Verified via docs

Claude Code

Agent Integration

Verified via docs

Codex CLI

Agent Integration

Verified via docs

Quick start

$ curl -fsSL https://raw.githubusercontent.com/vshulcz/deja-vu/main/install.sh | sh

Use cases

↳Reusing solutions from past agent conversations to avoid re-debugging or re-implementing.

↳Debugging complex issues by accessing comprehensive past debugging sessions and design decisions.

↳Maintaining persistent context across different AI agents and development machines.

↳Sharing sanitized session digests with colleagues for onboarding or knowledge transfer.

↳Understanding file change history and decision-making via session blame.

Alternatives

⚡Top VS:

VSprojectmem★870→VSWaggle-mcp★44→VSremem★34→VSMCP-memento★11→

projectmem★ 870

Local-first AI coding memory for AI agents. Records issues, attempts, fixes and decisions, then warns your agent before it repeats an approach that already failed. Native MCP server for Claude Code, Cursor, Antigravity and Codex. 100% local, no cloud, no telemetry. MIT.

Features

Why choose it

Trade-offs

Maintenance check

Most discussed open issues

Compatibility

Quick start

Use cases

Alternatives

Comments

Features

Why choose it

Trade-offs

Maintenance check

Most discussed open issues

Compatibility

Quick start

Use cases

Alternatives

Comments