100% local · source-available · zero dependencies

Your team shares a codebase.
Now your AI shares a memory.

Every line your AI tools write traces back to a prompt and a reason. MemBridge keeps that memory on your machine and shares it across Claude Code, Codex — and your whole team.

A tiny background daemon. Node 18+. Nothing leaves your machine.

~/cartful
The demo

Two tools. One memory.

Claude Code makes a decision. MemBridge distills it and amends the memory already living in CLAUDE.md and AGENTS.md — files your tools read on every start. Codex picks it up. Watch:

Claude Code · Sam · ~/cartful
watching session…
CLAUDE.mdamended
# cartful — dev notes <!-- membridge --> Cart drawer: optimistic. (Maya, Jul 15) Retries: backoff, cap 3 — gateway 10s limit. (Sam, Jul 17) <!-- /membridge -->
AGENTS.mdamended
# Agent instructions <!-- membridge --> Cart drawer: optimistic. (Maya, Jul 15) Retries: backoff, cap 3 — gateway 10s limit. (Sam, Jul 17) <!-- /membridge -->
Codex · Maya · next morning
The app

All of it, without the terminal.

The same memory, as a local dashboard: your projects, a plain-English feed of what your AI tools did, and every prompt behind every change. Built for people who ship with AI but don't live in a CLI — and for teammates catching up on each other's sessions.

MemBridge
Synced · 2m ago

Projects

All caught up across your projects — click one to open it

Ships with the daemon — membridge dashboard opens it. Menu-bar app on macOS.

Team sync

Your whole team's memory. Not just yours.

Share a project and every teammate's sessions land in the same feed — the summary up top, the exact prompts one click down. Stop re-asking what Maya already answered.

Invite-only

Projects are private by default. Share one, and only the people you invite see its memory — pause or leave any time.

End-to-end encrypted

Shared memory is encrypted on your machine before it syncs. The relay can't read your code, your prompts, or your summaries.

Summaries + exact prompts

Catch up from the feed: what changed in plain English — and the word-for-word prompts that produced it.

How it works

A background daemon, not a platform.

No accounts. No cloud. It reads the session logs your tools already keep, and writes to the context files they already read.

01 — WATCH

Tail the session logs

membridge start watches Claude Code and Codex transcripts — read-only, incremental, byte offsets only. Any tool that logs JSONL plugs in with a config entry. No code.

02 — DISTILL

Build a per-project memory

Recent asks, files touched, and the decisions behind them become a short digest. Secrets (sk-…, AKIA…, ghp_…) are scrubbed before a single byte is written.

03 — INJECT

Write where tools already read

The digest lands between <!-- membridge --> markers in CLAUDE.md and AGENTS.md. Your own notes stay byte-for-byte intact — and membridge remove restores everything exactly.

Privacy

Local is not a feature. It's the architecture.

127.0.0.1 only

The daemon binds to localhost. Solo use makes zero network calls — no telemetry, no accounts. Team sync is opt-in and end-to-end encrypted.

Zero dependencies

The core is plain Node. Nothing to audit but the code itself — and it's all there on GitHub.

Secrets redacted

Common key shapes are scrubbed by default before anything is written to a context file. Add your own patterns in config.

Fully reversible

MemBridge only ever touches the content between its own markers. One command strips every trace and restores your files.

Get it

Running in a minute.

install
$ git clone https://github.com/mmelika/membridge $ cd membridge $ node bin/membridge.js start

Node 18+ · macOS, Linux, Windows · npm package coming soon

Follow the build.

The npm release, the signed macOS app, and membridge why for every repo are landing soon. One email when they do — that's it.

No spam. A few emails a year. Unsubscribe = reply "stop".