Memoryvshulcz/deja-vu1.5k
deja-vu: local memory shared across coding agents
A single Go binary that indexes saved sessions from 43 coding agents on your own machine, then offers search, cross-agent handoff and recovery of overwritten code. DeepSeek Harness is one of the supported harnesses.
Install
dsh plugin --profile web add dsh-dejaThe --profile value names the profile the plugin is added to. Change it if you run a different one.
At a glance
- Cross-agent session search
- Session handoff
- Overwritten code recovery
- Local index
- Repository-level memory
What does deja-vu do?
deja-vu turns the session transcripts that coding agents already write to disk into one searchable local memory. Its command-line tool, deja, indexes those files so that any connected agent, or you in a terminal, can look up what was decided, run or edited before.
It is not specific to DeepSeek Harness. The project covers 43 agents, among them Claude Code, Codex, Cursor, Gemini CLI and OpenCode, with DSH as one entry on the list. Because the index is built from history that already exists, earlier sessions become searchable right after the first indexing pass, and a fix made in one agent can be recalled from another.
The same index supports two further jobs: passing an unfinished session to a different agent, and getting back code that an agent overwrote before it was committed.
Core features
- Search across agents: Running
dejawith no arguments opens a terminal search over every agent's history, with filters for agent, time range and project.deja viewexports the store to a single local HTML file. - Handoff:
deja handoffpasses the active session's goal, decisions, touched files and open work to another agent. - Restore and blame: The index records opened files, commands with exit codes and exact edit spans, so
deja restorecan recover overwritten code anddeja blamecan trace which session wrote a line. - Automatic recall: Through hooks and a single MCP tool, earlier sessions are surfaced at session start, before an edit or command, and when a command fails.
- Repository memory:
deja repo addwrites the decisions and dead ends you choose into.deja/memory.jsonl, so a teammate's agent can recall them after cloning. - Credential redaction: Keys, tokens and passwords are replaced with a redaction marker while the index is built. The original transcripts are left as they are;
deja secretslists the affected sessions.
How to use deja-vu
- Install the binary. On macOS or Linux the README uses an install script:
curl -fsSL https://raw.githubusercontent.com/vshulcz/deja-vu/main/install.sh | sh
The script does not run on Windows; use Scoop or winget there:
scoop install deja-vu
winget install vshulcz.deja-vu
- Wire it into the agents found on your machine. This sets up MCP recall and hooks and builds the first index:
deja install --auto
- If you prefer to install from the DSH plugin system, the README lists the npm package
dsh-deja:
dsh plugin --profile web add dsh-deja
- Start a new agent session, or query the history yourself:
deja "jwt refresh token"
deja handoff --to codex
deja restore internal/auth/jwt.go
Questions about deja-vu
How do I install deja-vu?
Run dsh plugin --profile web add dsh-deja in a terminal. The --profile value names the profile the plugin is added to, so change it if you run a different one.
Is deja-vu free to use?
The source is public at vshulcz/deja-vu under the MIT license. Check that license if you plan to use it commercially.
Is deja-vu still maintained?
The last code push was on Nov 11, 2026, and the latest release is v0.24.1. We last refreshed this data on Nov 12, 2026. DeepSeek Harness changes quickly, so a plugin untouched for months may no longer load.
Has Plugin DSH tested deja-vu?
No. It is listed from its public repository and we have not installed it. Read the source and recent issues before you run it.
Is deja-vu built only for DeepSeek Harness?
No. It is a general tool that reads the session stores of 43 coding agents, including Claude Code, Codex and Cursor. DeepSeek Harness is one supported harness, and the README lists a dedicated npm package for it, dsh-deja.
Which deja-vu features work in DeepSeek Harness?
The README's support table marks MCP recall, automatic recall, the skill, the command and handoff as available for DeepSeek Harness. Resume is marked as not supported, and zstd is listed as a requirement.
Does deja-vu send my session history over the network?
Indexing and search run locally with no LLM calls. According to the README, the network is used only by deja update, deja sync ssh, version checks and an embedding endpoint if you configure one; setting DEJA_OFFLINE=1 turns off the daily release check.