Nothing you did with AI was ever lost. It just wasn't searchable.

Fig. 1 — two commands and a question

$ npx chat-recall init
  index what is already on disk, wire up MCP

$ chat-recall search "that auth bug"
  every session, every tool, one ranked list

> what was I working on yesterday?
  the assistant answers from the index

Your data stays yours

Schedule 1 — process stages
No.	Stage	What happens	Where
01	The tools write	Each tool already logs every session on its own, in its own format and folder.	your disk
02	The CLI redacts	Credentials get masked before anything is sent. The last four characters survive; the rest of the key doesn't.	your machine
03	Postgres indexes	Everything turns into one searchable history.	your server, or ours
04	MCP answers	The assistant queries the index itself, so recall happens right inside the conversation.	back in the editor

Stages 01–02 run on your hardwareThe raw transcript never ships anywhere

Proactive history. Always up to speed.

Your agent autonomously reviews its past activity—no prompting required.

recall_smart_resume

A decision made in Codex reaches Claude Code

One index. Five tools. Your history stays yours—never locked into any single assistant.

recall_memory_search

Four things you get back

Per project, read back from your own sessions.

Fig. 2 — one project, read backFindings · coupling · decisions

A project view: a health score, ranked code findings naming the source files they came from, a coupling map of the codebase, and a knowledge graph of the decisions and tools drawn from past sessions. — Per project: what depends on what, what is risky, and what was decided. Top of the screen; it continues below.

Pick up right where you left off

Reopen unfinished work without explaining any of it again.

Trace any past decision

What was decided, when it was locked in, and when the facts changed.

Avoid redoing work

You get warned before you redo something you already finished.

Find leaked keys

Passwords and keys found in your own history, checked to see whether they still work.

What a task card carries

Every card carries the session that did the work. Open it and read the actual conversation.

Fig. 3 — session behind a cardOutcome · files · commits

One session's analysis: 24 messages, 8 files changed, 10 tool calls, a context-window bar, a per-prompt sentiment arc and marker tallies. — What was finished, which files it changed, and where the assistant ran out of room. Top of the screen; it continues below.

Reopen a task and your assistant already knows the story

It picks up from the conversation that did the work, with the files and the reasoning already in hand.

The answer comes back inside the assistant

Conversations, plans, task lists, notes, and shell history, all searched from one place.