recall-memory

recall

Transparent cross-chat memory for AI coding sessions. Every conversation is saved as a clean, readable markdown note you own, secrets scrubbed, and your recent memories are loaded back into each new chat. No black-box database. No cloud. Your memory, in your files.

npx recall-memory install

That's it. From then on, when you open a new Claude Code chat, it remembers the last ones.

The problem

You have a great session with Claude Code or Cursor. You solve something hard. Then you open a new chat and it remembers nothing. Every "memory" tool either lives in a database you can't read, or ships your conversations to someone's cloud. recall does neither.

How it works

recall uses Claude Code's native hooks:

• When a chat ends → it reads the transcript, writes a clean summary note (what you worked on, files touched, where you left off), and scrubs any secrets before saving.
• When you start a new chat → it loads your recent memories back into context.

Your memories are plain markdown in ~/.recall/memories/ — readable, greppable, and yours. Point RECALL_DIR at your Obsidian vault and they show up in your notes:

export RECALL_DIR="$HOME/vault/chats"

Commands

recall install          # wire into Claude Code (Stop + SessionStart hooks)
recall list             # list what it remembers
recall load             # print recent memories (what gets injected into new chats)
recall save <file.jsonl># save one transcript by hand
recall uninstall        # remove the hooks

Why it's different

	recall	cloud memory tools
Where your data lives	your own markdown files	their database
Can you read it?	yes, it's plain text	no
Leaves your machine?	never	yes
Secrets in your chats	scrubbed before saving	stored as-is

Secret-safe by design

Developers paste API keys into chats while debugging. recall redacts AWS/Stripe/GitHub/OpenAI/Anthropic keys, JWTs, and DB connection strings before anything is written, so your memory never becomes a place secrets leak from.

MIT. Built by Harrison Songolo.