MemoSync: agent memory under your control
MemoSync
A local workbench for Claude Code, Codex, Pi and OpenCode: memory changes wait for your review, then an audit follows each turn
- 178 ★ on the repository
- Last commit
Extended access
Why this level
- It runs agents in your projects and stores API keys in `~/.memosync/data/credentials.json`
- In Automatic mode safe memory changes apply without your review, with Revert only
Install
- Works in
Manual install
git clone https://github.com/SLEEPYBQ/MemoSync.git && cd MemoSync && bun install && bun run build && bun run startNeeds Bun 1.3.5 or newer. The workbench opens at `http://localhost:3210`.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- It runs agents in your projects and stores API keys in `~/.memosync/data/credentials.json`
- In Automatic mode safe memory changes apply without your review, with Revert only
This is third-party code. Review the repository files before installing.
What it does
Coding agents keep memory: `CLAUDE.md` files, notes, lessons from earlier sessions. You rarely see what it holds, what the agent took into a task and whether it followed it. MemoSync puts memory on screen. It is a local workbench in the browser at `http://localhost:3210` that runs Claude Code, Codex, Pi or OpenCode inside it. Each turn has four steps. Before the turn the workbench shows candidate memory changes: new candidates from the last exchange, rule transfers from other projects, fixes for duplicates and stale items. Then you choose the working memory for the turn. While the agent works, replies cite items such as `M-07`, and a Stop button on a citation interrupts the turn and saves your correction. After the turn a separate check rates every item: violated, shaped the turn, not applicable or no visible effect. Memory is split into scopes: personal, project and session. The Memory Board shows the library as a graph or as columns, and an existing `CLAUDE.md` is imported as candidates for review. In Automatic mode safe changes apply on their own, with Revert.
Who it is for. Developers who work with one agent for long stretches and want to see and edit what it remembers about a project instead of trusting accumulated notes blindly.
Good fit when
- The agent repeats an old mistake and you want to find the memory item that nudges it
- Rules from one project suit another and should be carried over in generalized form
- `CLAUDE.md` has grown large and there is no way to check whether it is followed
- Before a turn you want to confirm which memory items the agent will get
- Quick questions go without review: switch to Automatic and watch only for violations
Not a fit when
- The agent's memory suits you and you do not need a detailed check of every turn
- You need an environment without a local server on a port: MemoSync runs one on `localhost:3210`
- The audit must be infallible: a model gives the rating and it can be wrong
Example request
Add validation to the sign-in form. Before the turn show which memory items you will use, and after the turn report which of them were violated.Limitations
Installing from source needs Bun 1.3.5 or newer and a sign-in to Claude Code or Codex, or an API key from any provider. The Mac app is not notarized, so macOS blocks it on first launch until you allow it in Privacy & Security.
The app and a source install share `~/.memosync/data`, and only one runs at a time. The app contains no agents: Claude Code and Codex must be installed beforehand.
Citations show what the agent says it used and do not prove it, which is why the audit exists. The audit rating is a model judgment. The repository's last commit is from November 2026.
How to disable. Stop the server and delete the project directory. In Settings, Memory, Use Memory turns memory off everywhere while the library stays on the board. Data lives in `~/.memosync/data`.
FAQ
How do I run MemoSync from source?
Clone the repository and run `bun install`, `bun run build`, `bun run start`. The workbench opens at `http://localhost:3210`. The `--port` flag changes the port and `--no-open` stops the browser from opening automatically.
Which agents run inside MemoSync?
Claude Code, Codex, Pi and OpenCode 1.17.4 or newer. Claude Code and Codex come from your own install, Pi is built into the app. Access goes through your sign-in or an API connection.
Does MemoSync change my Claude Code settings?
No. Engines on an API connection run in a separate profile under `~/.memosync/data/cli-profiles`, so your `~/.claude` and `~/.codex` are never read or changed. Keys are stored in `~/.memosync/data/credentials.json`.
What does the audit show after a turn?
A separate pass reads the whole exchange and rates each selected item: violated, shaped the turn, not applicable or no visible effect. Each rating names its source, so a violation the agent never reported still shows up.
How do I bring in an existing CLAUDE.md?
On the Memory Board, Memory files imports a `CLAUDE.md` file as candidates for review, and nothing reaches the store until you decide. A Markdown copy for other tools is kept there as well.
Discussion
No questions or reviews yet.
Related
CLI toolsShared local memory for AI agents: Claude Code, Codex, Cursor and others find the agreements of past sessions in it and record new ones
Persistent memory for coding agents: insights are stored as markdown cards and recalled at the start of the next session.
A long-term memory plugin that keeps decisions, preferences and task state across sessions and injects only what a query needs
Long-term memory for agents: it distills knowledge from session history and brings it back in new tasks, one store shared across Claude Code, Codex, Cursor and
MemoMind MCP gives Claude Code project memory, with stored facts, events and conclusions visible in a local dashboard.
Local memory for agents: stores conversations and project files verbatim and retrieves them with semantic search over MCP
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| MemoSync: agent memory under your control (this card) | no | 178 | |
| Claude Code Memory Setup: Obsidian-based memory | no | 1,059 | memory file setup without a review panel or audit |
| Supermemory Skills | yes | 116 | cloud memory as a service, without manual review of each item |
README in short
The README follows the four steps of a turn with illustrations: reviewing new memory, choosing working memory, following replies and the audit. The Memory Board, Automatic mode, connecting engines, settings and running on another port come next.