localmem-app

mcp
Security Audit
Warn
Health Warn
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Local, private memory for AI agents — one vault on your Mac, shared by Claude Code, Cursor, Codex and more over MCP

README.md

Localmem

Persistent memory for your AI agents.

Localmem is private, file-based memory that lets your AI agents recall context
across every conversation and project — over the Model Context Protocol,
all on your machine. Open-source core, macOS-native.

localmem.ai

Core license: Apache 2.0
App license: FSL-1.1-ALv2
CI
Platform: macOS

Status: early and under active development. macOS-only.
Signed & notarized builds are on the releases page.


What it is

AI agents forget everything the moment a conversation ends. Localmem gives them
a memory that stays: a local SQLite store your agents read from and write to
over MCP, so a preference you state in one session — or a decision you make in
one project — is there in every session after, across every project.

Memory is user-level and cross-project. A fact stored in one repo's session
is visible from every other repo's session. Nothing leaves your Mac: no account,
no server, no cloud.

Three surfaces sit on top of one shared LocalmemCore engine, so the app, the
CLI, and the MCP server all read and write the same database with identical
search, permissions, and audit behavior:

Binary Role
localmem-app The SwiftUI vault app — browse memory, manage agent access, review the audit log
localmem The command-line tool — setup, add, search, list, and more
localmem-mcp The MCP server your AI clients actually talk to

Features

  • Local and file-based. Memory lives in a SQLite database on your machine.
    No account, no backend, no sync required.
  • Cross-project by design. One memory store, shared across every repo and
    every agent session.
  • Full-text search over every memory via SQLite FTS5.
  • Per-agent access control. Grant or revoke each connected agent's access to
    individual memories.
  • Audit log. Every read, search, and write is attributed to the agent that
    made it and recorded so you can see what was accessed.
  • Import from your files. Pick Text, Markdown, or PDF files and Localmem
    extracts the key facts into memories — on-device via Apple Intelligence when
    available, otherwise through a CLI agent you already use (Claude Code or
    Codex). Every candidate fact is then verified against the source document
    before anything is stored, and each file shows its "N extracted → M kept"
    tally. You choose every file; nothing is scanned automatically.
  • One-command setup. localmem setup registers the MCP server with every
    installed client. Your client asks you to approve Localmem's tools the first
    time an agent uses them — approve it once. (Pass --preauthorize to skip that
    prompt if you'd rather auto-approve.)

MCP tools

Agents interact with Localmem through four tools:

  • memory_store — persist a new fact, preference, or decision
  • memory_search — full-text search over stored memories
  • memory_recent — most-recent-first listing
  • memory_update — replace fields on an existing memory

Supported clients

Setup auto-detects and registers Localmem with:

  • Claude Code
  • Claude Desktop
  • Cursor
  • Codex
  • Antigravity

Getting started

Localmem is macOS-only (macOS 26+, Apple Silicon & Intel).

Download: grab the signed, notarized DMG from the releases page
(or localmem.ai), drag Localmem to Applications, and launch. The setup
wizard connects your installed AI clients and can add the localmem CLI to
your PATH in one click.

Or build from source:

git clone https://github.com/localmemai/localmem-app.git
cd localmem-app
swift build

# Register the MCP server with every installed AI client
swift run localmem setup

Once setup finishes, restart your AI clients and they'll have Localmem
available. Try telling one "remember that I prefer flat whites" — then ask a
fresh session in another project what you like to drink.

Homebrew and a curl installer are planned as additional channels.

CLI

The localmem command is the pro-user and debugging surface:

Command Description
localmem setup Register the MCP server with all installed clients
localmem add Add a memory
localmem search Full-text search
localmem list List recent memories
localmem show Show a single memory
localmem update Edit a memory
localmem delete Remove a memory
localmem status Show store and registration status
localmem path Print the database path

Run localmem --help for the full command set.

Architecture

Localmem is a single Swift package. LocalmemCore owns memory CRUD, search,
per-agent permissions, and audit logging; the app, CLI, and MCP server are thin
surfaces over it. Storage is SQLite via GRDB with an FTS5 index for
search. The MCP server is built on the official Swift MCP SDK.

The full architecture, data model, security model, access control, app UI, and
distribution plan live in the technical design.

Contributing

Contributions are welcome — see CONTRIBUTING.md.

License

Localmem is open-core — see LICENSING.md for the full picture:

  • Core, CLI, and MCP serverApache 2.0. Everything that
    touches your memories is fully open source.
  • macOS app (Sources/localmem-app) —
    FSL-1.1-ALv2: source-available so you can
    verify everything stays on your device; free to use and contribute to; no
    competing use. Each release converts to Apache 2.0 after two years.

Reviews (0)

No results found