OMEGA

AI agents that remember, coordinate, and learn. All on your machine. Your agent's brain shouldn't live on someone else's server.

The Problem

AI coding agents are stateless. Every new session starts from zero. And the "solutions" want you to send your codebase context to their cloud.

• Context loss. Agents forget every decision, preference, and architectural choice between sessions. Developers spend 10-30 minutes per session re-explaining context that was already established.
• Repeated mistakes. Without learning from past sessions, agents make the same errors over and over. They don't remember what worked, what failed, or why a particular approach was chosen.
• Cloud memory = someone else's database. Services like Mem0 require API keys and send your data to their servers. When they change pricing, get acquired, or go down, your agent's accumulated intelligence disappears.

OMEGA solves this. Memory, coordination, and learning that runs entirely on your machine. No cloud. No API keys. No vendor lock-in.

Quick Install

pip install omega-memory[server]    # Full install (memory + MCP server)
omega setup                         # Downloads model, registers MCP, installs hooks
omega doctor                        # Verify everything works

If you only need OMEGA as a Python library for scripts, CI/CD, or automation:

pip install omega-memory    # Core only, no MCP server process

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 (~50 MB lighter, no background process). Hooks still work:

omega setup --hooks-only    # Auto-capture + memory surfacing, no MCP server (~600MB RAM saved)

From Source

git clone https://github.com/omega-memory/omega.git
cd omega
pip install -e ".[server,dev]"
omega setup

omega setup will:

1. Create ~/.omega/ directory
2. Download the ONNX embedding model (~90 MB) to ~/.cache/omega/models/
3. Register omega-memory as an MCP server with Claude Code
4. Install session hooks into ~/.claude/settings.json
5. Add an OMEGA block to ~/.claude/CLAUDE.md

60-Second Quickstart

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.

Key Features

• Memory & Learning — Stores decisions, lessons, error patterns, and preferences with semantic search. Claude recalls what matters without you re-explaining everything each session. 25 memory tools including compaction, consolidation, timeline, graph traversal, and context virtualization (checkpoint/resume).

• Multi-Agent Coordination (omega-pro) — File and branch locking, session management, task queues with dependencies, intent broadcasting, and agent-to-agent messaging. 29 coordination tools that prevent agents from overwriting 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.

How OMEGA Compares

Architecture

               ┌─────────────────────┐
               │    Claude Code       │
               │  (or any MCP host)   │
               └──────────┬──────────┘
                          │ stdio/MCP
               ┌──────────▼──────────┐
               │   OMEGA MCP Server   │
               │   25 core 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 via omega-pro and register into the same server process. No separate daemons, no microservices.

MCP Tools Reference

OMEGA runs as an MCP server inside Claude Code. The core package provides 25 memory tools. omega-pro adds coordination, routing, entity, knowledge, and profile tools.

Memory (25 tools)

Additional tools with omega-pro

CLI

Hooks (7 processes, 11 handlers)

All hooks dispatch via fast_hook.py → daemon UDS socket, with fail-open semantics.

With omega-pro, additional coordination handlers register automatically: session lifecycle, file/branch claim guards, heartbeat, and git push guards.

Storage

Search Pipeline

1. Vector similarity via sqlite-vec (cosine distance, 384-dim bge-small-en-v1.5)
2. Full-text search via FTS5 (fast keyword matching)
3. Type-weighted scoring (decisions/lessons weighted 2x)
4. Contextual re-ranking (boosts by tag, project, and content match)
5. Deduplication at query time

Memory Lifecycle

• Dedup: SHA256 hash (exact) + embedding similarity 0.85+ (semantic) + Jaccard per-type
• Evolution: Similar content (55-95%) appends new insights to existing memories
• TTL: Session summaries expire after 1 day, lessons/preferences are permanent
• Auto-relate: Creates related edges (similarity >= 0.45) to top-3 similar memories
• Compaction: Clusters and summarizes related memories

Memory Footprint

• Startup: ~31 MB RSS
• After first query (ONNX model loaded): ~337 MB RSS
• Database: ~10.5 MB for ~242 memories

What Gets Modified

omega 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 <!-- OMEGA:BEGIN --> block

All changes are idempotent.

Troubleshooting

omega doctor shows FAIL on import:

• Ensure pip install -e ".[server]" from the repo root
• Check python3 -c "import omega" works

MCP server fails to start:

• Run pip install omega-memory[server] (the [server] extra includes the MCP package)

MCP server not registered:

claude mcp add omega-memory -- python3 -m omega.server.mcp_server

Hooks not firing:

• Check ~/.claude/settings.json has OMEGA hook entries
• Check ~/.omega/hooks.log for errors

Development

pip install -e ".[server,dev]"
pytest tests/                # 2198+ tests
ruff check src/              # Lint

Uninstall

claude mcp remove omega-memory
rm -rf ~/.omega ~/.cache/omega
pip uninstall omega-memory

Manually remove OMEGA entries from ~/.claude/settings.json and the <!-- OMEGA:BEGIN --> block from ~/.claude/CLAUDE.md.

Contributing

• Contributing Guide
• Security Policy
• Changelog
• Report a Bug

License

Apache-2.0. See LICENSE.