open-memex

Persistent memory for AI coding agents — on your machine, in plain Markdown, shared by every tool you code with.

GitHub repository · npm package · Documentation · Releases

Every AI coding session starts from zero: you re-explain the project, the agent rediscovers the same gotchas, and yesterday’s decisions vanish when the chat ends. open-memex gives your agents a memory that survives the session.

Say “remember: we deploy on Fridays” once. Next week, Copilot, Cursor, opencode, or Claude Code already knows — because they read and write the same local memory on your machine.

Get started in about a minute

Requirements: Node.js 22.14 or newer.

npm install -g open-memex
open-memex init
open-memex add "We deploy only before noon on Fridays" --type decision

Then ask your agent what it remembers. open-memex init auto-detects supported editors and wires up the integration for you.

Why open-memex?

• One memory, every agent. Developers rarely use only one AI tool. open-memex avoids separate memory silos by giving the tools on one machine a shared local store.
• Your files, your rules. Memories are plain Markdown files you can read, edit, search with normal tools, back up, or delete. A rebuildable SQLite FTS5 index makes them fast to retrieve.
• Local-first by design. No account, no cloud service, no telemetry. Personal memories never leave the machine unless you explicitly export or share something.
• Human review stays in the loop. Capture can be automatic or agent-assisted, but memory remains inspectable and correctable — hide it, supersede it, forget it, or review it before it is shared.
• Teams share through Git. Project knowledge can move from a personal draft to a reviewed project memory through the normal branch / pull-request workflow instead of a separate sync service.

How it works

1. Capture — Save a decision, constraint, preference, lesson, or project fact through your agent, the CLI, or supported keyword triggers.
2. Store — open-memex writes a structured Markdown memory and updates its local SQLite keyword index.
3. Recall — In a later session, the agent can search the memory or receive relevant context automatically where the host supports first-turn injection.
4. Review and share — Personal memories stay local. Project memories can be proposed, reviewed, approved, and published through Git.

Works with the tools you already use

open-memex provides a native opencode plugin and a standard MCP server used by tools including:

• VS Code Copilot
• Cursor
• Claude Code
• Visual Studio
• opencode

The same memory layer is underneath each integration, so a constraint captured in one editor can be respected in another.

Built for knowledge that should outlive a chat

Use open-memex for the long tail of engineering knowledge that rarely belongs in a formal design document:

• deployment rules and environment quirks
• API limits and integration constraints
• architecture decisions and the reasons behind them
• recurring failure modes and their fixes
• team conventions and review expectations
• your own working preferences

Privacy and safety

• Memories live locally; there is no open-memex cloud.
• Common API keys and tokens are masked before a memory is saved.
• Text wrapped in <private>…</private> is stripped before saving.
• personal scope never syncs. Project sharing is explicit.
• Run open-memex doctor to check the installation, scope resolution, storage, editor wiring, and SQLite driver.

Project status

open-memex is open source under the Apache-2.0 license. The current stable release is 0.7.1. Start with the GitHub README for the full guide, or open a question in GitHub Discussions.

Created by Mr Sun.