OMEGA
Cross-model memory for AI agents. Local-first. Works with Claude, GPT, Gemini, Cursor, Claw Code, and any MCP client. Your agent's brain shouldn't live on someone else's server, or be locked to one provider.
接入你的智能体
發布者自己的指令。不是 PlugMCP 端點。
用我們的金鑰與日誌託管一份副本。
uvx omega-memory==1.3.0Claude Code 沒有本機指令的安裝連結。複製下方指令並在終端機中執行。
claude mcp add omega-memory-omega-memory -- uvx omega-memory==1.3.0在終端機中執行。
{ "mcpServers": { "omega-memory-omega-memory": { "command": "uvx", "args": [ "omega-memory==1.3.0" ] } }}claude_desktop_config.json
遠端 URL 還是本機指令? 兩種安裝方式的差異
AI coding agents are stateless. Every new session starts from zero. The "solutions" either lock you into one model provider or send your codebase context to their cloud.
OMEGA solves this. Memory, coordination, and learning that runs entirely on your machine. Works with every major LLM and coding agent. No cloud. No API keys. No vendor lock-in.
OMEGA needs Python 3.11 or newer. Check with python3 --version: macOS ships Python 3.9, which is too old. If pip answers No matching distribution found for omega-memory, that is the reason; install a newer Python first (for example brew install [email protected], or from python.org).
Install OMEGA into its own environment. Homebrew's Python refuses a plain pip install (PEP 668), and an isolated install keeps OMEGA's dependencies away from your projects. Pick one:
# pipx
pipx install "omega-memory[server]"
# or uv
uv tool install "omega-memory[server]"
# or a virtual environment
python3.12 -m venv ~/.venvs/omega # any Python 3.11 or newer
~/.venvs/omega/bin/pip install "omega-memory[server]"
export PATH="$HOME/.venvs/omega/bin:$PATH"
Keep the quotes around "omega-memory[server]": zsh, the default macOS shell, treats bare square brackets as a pattern and fails. The [server] extra installs the MCP server, which also runs the hook daemon; without it Claude Code gets no OMEGA tools and no hooks.
Then:
omega setup # downloads the embedding model, registers the MCP server, installs hooks
omega doctor # checks every step, and names the fix for anything that failed
omega setup --dry-run shows what setup would change without writing or downloading anything.
omega setup --client claude-desktop
This registers OMEGA as an MCP server in Claude Desktop's config. Restart Claude Desktop to activate.
omega setup --client cursor # or: windsurf, cline, codex, antigravity
Codex and Antigravity get their config file written; for the others setup prints the JSON block to paste. For any other MCP client, omega setup --client venv prints the command and arguments to use. Hooks (automatic capture and surfacing) are available with Claude Code only.
If you only need OMEGA as a Python library for scripts, CI/CD, or automation, install it into your project's environment:
pip install omega-memory # Core only, no MCP server
from omega import store, query, remember
store("Always use TypeScript strict mode", "user_preference")
results = query("TypeScript preferences")
This gives you the full storage and retrieval API without running an MCP server. Hooks need the MCP server (the hook daemon runs inside it), so a library-only install has no automatic capture or surfacing.
git clone https://github.com/omega-memory/omega-memory.git
cd omega-memory
python3.12 -m venv .venv && source .venv/bin/activate
pip install -e ".[server,dev]"
omega setup
omega setup will:
~/.omega/ (or $OMEGA_HOME if you set it)~/.cache/omega/models/omega-memory as an MCP server (Claude Code auto-detected, or specify --client)~/.claude/settings.json~/.claude/CLAUDE.mdIf a step fails, the summary marks it [FAIL] with the reason and setup exits with an error.
OMEGA works through natural language — no API calls, no configuration. Just talk to Claude.
1. Tell Claude to remember something:
"Remember that the auth system uses JWT tokens, not session cookies"
Claude stores this as a permanent memory with semantic embeddings.
2. Close the session. Open a new one.
3. Ask about it:
"What did I decide about authentication?"
OMEGA surfaces the relevant memory automatically:
Found 1 relevant memory:
[decision] "The auth system uses JWT tokens, not session cookies"
Stored 2 days ago | accessed 3 times
That's it. Memories persist across sessions, accumulate over time, and are surfaced automatically when relevant — even if you don't explicitly ask.
Memory & Learning — Stores decisions, lessons, error patterns, and preferences with semantic search. Claude recalls what matters without you re-explaining everything each session. Tools cover compaction, consolidation, timeline, graph traversal, and checkpoint/resume.
Multi-Agent Coordination (omega-pro) — File and branch locking, session management, task queues with dependencies, intent broadcasting, and agent-to-agent messaging, so agents don't overwrite each other's work.
Intelligent LLM Routing (omega-pro) — Classifies tasks and routes to the optimal model. Coding → Claude Sonnet. Quick edit → Llama 8b at 1/60th the cost. 1M token context → Gemini Flash. 5 providers, 4 priority modes, sub-2ms intent classification.
Knowledge Base (omega-pro) — Ingest PDFs, markdown, web pages, and text files into a searchable knowledge base with semantic chunking.
Entity Registry (omega-pro) — Multi-entity corporate memory with relationships, hierarchies, and entity-scoped memories/profiles/documents.
Secure Profile (omega-pro) — AES-256 encrypted personal data storage with macOS Keychain integration.
Anthropic's Memory Tool stores data client-side but requires Claude API calls for all memory operations. OMEGA runs entirely on-device, including embeddings (ONNX).
Anthropic Memory is for Anthropic. OMEGA is for everyone.
Claude Code · Cursor · Claw Code · Any MCP Client
│ │ │ │
└─────────┴─────┬───┴──────────────┘
│ stdio/MCP
┌───────────────▼─────────────┐
│ OMEGA MCP Server │
│ core memory tools │
└──┬──────────────────┘
│
┌────────▼──────────────┐
│ Core Memory Engine │
│ (semantic search, │
│ embeddings, graphs) │
└─────┬─────────────────┘
│
▼
┌──────────────────────────────────────┐
│ omega.db (SQLite) │
│ memories | edges | embeddings │
└──────────────────────────────────────┘
Single database, modular handlers. Optional modules (coordination, router, entity, knowledge, profile) are available with OMEGA Pro and register into the same server process. No separate daemons, no microservices.
OMEGA runs as an MCP server inside Claude Code. By default it shows the client five tools (omega_store, omega_welcome, omega_protocol, omega_tools, omega_call) to save context: omega_tools lists everything available and omega_call runs any of it. Set OMEGA_CONDENSED=0 in the server's environment to expose every tool directly.
OMEGA Pro adds coordination, routing, entity, knowledge base, and profile tools.
Core is free and open source. On the free tier, once the store holds 2,000 memories search switches from semantic search to a plain text match, and new memories stop being stored at 5,000. omega status shows where you are. OMEGA Pro removes both limits.
All hooks run fast_hook.py, which hands the work to the hook daemon inside the MCP server over ~/.omega/hook.sock. If no server is running, the hooks return at once without doing anything, except reply capture, which runs on its own; none of them blocks Claude Code.
Memory text the hooks inject is labelled as stored data, not instructions.
With OMEGA Pro, additional coordination handlers register automatically: session lifecycle, file/branch claim guards, heartbeat, and git push guards.
omega serve install runs one shared server under launchd instead of one per session (macOS). It listens on 127.0.0.1:8377, refuses requests whose Host or Origin is not local, and requires a bearer key kept in ~/.omega/mcp_api_key. omega serve migrate-config points Claude Code at the daemon and writes that key into its config; omega serve restore-config switches back.
related edges (similarity >= 0.45) to top-3 similar memoriesomega setup modifies these files outside ~/.omega/:
~/.claude.json — Adds omega-memory MCP server entry~/.claude/settings.json — Adds hook entries~/.claude/CLAUDE.md — Adds a managed `` blockAll changes are idempotent.
omega doctor shows FAIL on import:
pip install -e ".[server]" from the repo rootpython3 -c "import omega" worksMCP server fails to start:
pip install "omega-memory[server]" (or pipx install --force "omega-memory[server]")MCP server not registered:
omega setup again, or run the claude mcp add -s user omega-memory -- ... command that omega doctor prints: it names the Python from OMEGA's own environmentHooks not firing:
omega hooks doctor; omega hooks setup repairs missing or outdated entries~/.omega/hooks.log for errorspip install -e ".[server,dev]"
pytest tests/ # Test suite
ruff check src/ # Lint
claude mcp remove -s user omega-memory
omega serve uninstall # only if you installed the HTTP daemon
rm -rf ~/.omega ~/.cache/omega
pipx uninstall omega-memory # or: uv tool uninstall omega-memory, or delete the venv
Manually remove OMEGA entries from ~/.claude/settings.json and the `` block from ~/.claude/CLAUDE.md.
Apache-2.0. See LICENSE.
Persistent memory for AI coding agents
{ "mcpServers": { "omega-memory-omega-memory": { "command": "uvx", "args": [ "omega-memory==1.3.0" ] } }}工具
尚無工具清單。遠端伺服器會在首次健康檢查後取得工具清單。
常見問題
在本頁複製適用於 Cursor、Claude Desktop、Claude Code 或 VS Code 的 stdio 設定。它使用發布者的安裝方式,而不是 PlugMCP 代理伺服器。
這筆記錄未列出必要的環境變數。
Omega Memory 透過 stdio 在本機執行。
上次檢查時間:2026-10-26。 Persistent memory for AI coding agents