Skip to content

venkateshamatam/recall

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

24 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

recall

Cross-agent memory for AI coding assistants. One local SQLite database, every MCP-capable agent on your Mac reads and writes it.

If you switch between AI agents during a project, recall keeps what each one learns about you in a single place the others can read.

Install

curl -fsSL https://raw.githubusercontent.com/venkateshamatam/recall/main/install.sh | sh
recall init
recall install --all

Requires Node 20 or later. macOS only.

What gets installed

recall install --all does a few things.

It registers a local MCP server with Claude Code, Claude Desktop, Cursor, Windsurf, and Zed. The server exposes five tools: recall_save, recall_search, recall_list, recall_get, and recall_delete.

It adds a Claude Code SessionEnd hook so the last assistant turn from each session gets saved automatically and tagged with the current git project.

It wires the auto-inject layer. That's a UserPromptSubmit hook for Claude Code, a user rule at ~/.cursor/rules/recall.mdc, and a marker block in ~/.agents/AGENTS.md. The Claude Code hook prints relevant memories to stdout before each prompt, and Claude Code merges that into the model's context. Cursor and Claude Desktop fall back to the rule and the MCP tools.

Storage

~/.recall/db.sqlite        SQLite + sqlite-vec, plain file
~/.recall/models/          all-MiniLM-L6-v2 quantized, ~23 MB, downloaded once

Embeddings run on-device via transformers.js. The model downloads once on first use, and after that recall does not make network calls. To sync across machines, point ~/.recall at iCloud, Dropbox, or syncthing.

CLI

recall init                       set up ~/.recall/, download embedding model
recall install [--all|--agent X]  wire mcp + auto-inject + auto-capture
recall doctor                     check what's wired
recall setup-prompt               paste-into-any-agent installer prompt

recall add <text>                 save a memory
recall search <q> [--all]         semantic search (defaults to current project)
recall list [--all]               recent memories
recall context                    write ~/.recall/context-<project>.md for @-import
recall export [<file>]            json dump

recall server                     run the mcp server (called by mcp clients)
recall capture                    save piped transcript (used by SessionEnd hooks)
recall inject                     print recall context to stdout (used by UserPromptSubmit hooks)

Granular control

recall install --all --bare       MCP wiring only, skip auto-inject and auto-capture
recall auto install               just the auto-inject layer
recall auto uninstall             remove the auto-inject layer
recall hooks install --all        just the auto-capture layer
recall hooks uninstall --all      remove the auto-capture layer

Paste-into-any-agent install

recall setup-prompt | pbcopy

Paste the output into any agent that can run shell commands. It runs the install and verifies with recall doctor.

License

MIT.

Issues

https://github.com/venkateshamatam/recall/issues

About

One shared memory pool for every AI agent on your machine. Local-first MCP server.

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors