origin

https://github.com/user-attachments/assets/6407e96a-9b84-4506-b702-c4a5f8da2920

Origin runs quietly behind the AI tools you already use. It gives your AI a place to carry decisions, lessons, gotchas, and project context instead of rediscovering them in every new chat.

The daemon does the memory chores in the background: deduplicating, linking related ideas, compiling concepts, and keeping provenance attached.

Use optional desktop app to search, inspect, edit, and delete what Origin learned.

Status: Early preview for macOS Apple Silicon. Expect fast iteration and some sharp edges.

Quickstart

Platform: macOS Apple Silicon (M1+). Linux, Intel Mac, and Windows are not supported yet.

1. Use with Claude Code, Cursor, Codex, or another MCP client

For Claude Code, Cursor, Codex, Claude Desktop, Windsurf, Gemini CLI, or another client that accepts a JSON mcpServers entry, add:

{
  "mcpServers": {
    "origin": {
      "command": "npx",
      "args": ["-y", "origin-mcp"]
    }
  }
}

On first run, npx origin-mcp downloads the MCP binary from the origin-mcp repo. It connects to the local Origin daemon on 127.0.0.1:7878, started by the desktop app or the headless install below.

2. Add the desktop app

Use the desktop app when you want Origin in the menu bar, memory inspection/editing, and Remote Access for Claude.ai or ChatGPT web.

1. Download the .dmg from GitHub Releases.
2. Drag Origin into Applications.
3. Clear quarantine (the build is currently unsigned):

   sudo xattr -cr /Applications/Origin.app
   

4. Launch Origin. It runs from the menu bar and starts the local daemon on 127.0.0.1:7878.

If you already use the MCP config above, origin-mcp connects to the desktop app's daemon.

3. Claude.ai and ChatGPT web

Use Remote Access from the desktop app. Web clients do not use the local stdio npx config above.

4. Headless daemon only

Use this path for automation, servers, or no-GUI setups.

curl -fsSL https://raw.githubusercontent.com/7xuanlu/origin/main/install.sh | bash
export PATH="$HOME/.origin/bin:$PATH"
origin-server install
origin-server status

Why Origin?

AI work has a continuity problem. Most memory tools ask you to trust a cloud API, maintain a notes/wiki workflow, or move into another app. Origin takes a different shape: a local daemon, a desktop inspection lens, and MCP access from the AI tools you already use.

Your AI starts from scratch too often. Origin carries decisions, preferences, gotchas, and project context across chats, projects, and time.

Memory gets worse when nobody maintains it. Origin runs a background refinery that deduplicates captures, links related ideas, compiles concepts, and keeps provenance attached.

You need to see and correct what it learned. The desktop app lets you search, inspect, edit, and delete memories instead of trusting a black box.

Developers already have origin for code. Origin gives AI work a source of truth too: local, traceable, and readable by agents through MCP.

For people whose work spans projects, clients, and jobs. Your context should not disappear when a chat ends or when you switch gears.

Origin keeps the useful parts together:

• Capture: decisions, lessons, observations, gotchas, and project context.
• Refine: deduplicate, link, and compile memories in the background.
• Recall: relevant context through MCP when your AI needs it.
• Inspect: every memory stays editable and traceable to where it came from.

96% fewer tokens per query. Same cost as basic vector search, but 19% more relevant context. 168 tokens instead of 4,505 for full replay. Measured on LoCoMo (2,531 memories, 1,540 queries). Eval harness at crates/origin-core/src/eval/.

How Origin works

Use your AI tools normally. Origin runs in the background, keeps the useful parts, and makes them available when context is needed.

1. Install Origin. The desktop app lives in the macOS menu bar and keeps the local daemon running.
2. Connect your AI tools. Claude Code, Cursor, Codex, Claude Desktop, Windsurf, Gemini CLI, and other MCP clients connect through origin-mcp. Claude.ai and ChatGPT web use Remote Access.
3. Capture useful context. Agents save decisions, preferences, project facts, gotchas, and lessons while you work.
4. Refine in the background. Origin deduplicates captures, links related ideas, compiles concepts, and preserves where each memory came from.
5. Recall when needed. Retrieval combines vector search, full-text search, and knowledge graph signals without replaying full chat history.
6. Inspect and export. Use the desktop app to search, edit, delete, verify, and export what Origin learned.

Evaluation

Retrieval quality on standard long-memory benchmarks. Numbers come from BGE-Base-EN-v1.5-Q embeddings combined with FTS5 and Reciprocal Rank Fusion. Harness at crates/origin-core/src/eval/; update workflow in docs/eval.

Local by default

• Memories are stored locally at ~/Library/Application Support/origin/memorydb/origin_memory.db by default.
• The daemon listens on 127.0.0.1:7878; the desktop app and MCP clients call that local API.
• There is no cloud sync or telemetry by default. Remote access and external LLM endpoints are opt-in settings.
• On-device Qwen models download only when requested and use the hf-hub cache.
• Security reports: SECURITY.md.

Architecture

Origin is daemon-first. origin-server owns the local database, embeddings, refinery, knowledge graph, and HTTP API on 127.0.0.1:7878. The desktop app and MCP clients are thin clients over that daemon.

Stack: Rust, Tauri 2, libSQL, Tokio, FastEmbed, llama-cpp-2, Axum, React, Tailwind CSS. Full module map: CLAUDE.md.

Build from source

git clone https://github.com/7xuanlu/origin.git
cd origin
pnpm install

Single command builds the daemon, starts it, and launches the Tauri app with Vite:

pnpm dev:all

Or run daemon and app separately:

cargo run -p origin-server          # terminal 1
pnpm tauri dev                      # terminal 2

For local .dmg builds:

pnpm release            # builds daemon + app bundle
pnpm release:dmg                # wraps .app into DMG

First build takes several minutes while llama.cpp compiles for Metal.

Boundaries

• Not a chat UI. Keep using Claude, ChatGPT, Cursor, or your agent of choice.
• Not a notes app or Notion / Obsidian replacement. Markdown export exists so you can read the artifact anywhere.
• Not a memory infrastructure SDK. Origin is meant for people using AI, not as a backend for other apps.
• macOS Apple Silicon only for now.
• Best for work that spans sessions, projects, and weeks. One-off chats may not need it.

Contributing

Bug fixes, eval cases, docs, and features are welcome. Start with CONTRIBUTING.md. Please also read the Code of Conduct. Architecture is in CLAUDE.md.

License

• origin-types, origin-core, origin-server: Apache-2.0
• Desktop app (app/) and root frontend UI: AGPL-3.0-only
• origin-mcp (MCP server, npm package, Homebrew): MIT

The split keeps the data layer permissively licensed for downstream tools while the shipped desktop app stays AGPL.

Acknowledgments

Adjacent work shaping this space:

• Andrej Karpathy's note on the LLM-wiki pattern, parallel work in this space.
• Claude Code's MEMORY.md, the simplest version of the idea, and the one Origin aims to cooperate with.
• PAI, claude-memory-compiler, Palinode: different shapes of the same direction.