This is an early release preview. You may encounter bugs.
Omega Memory logo
Unclaimed

Tool coding memory

Omega Memory

Persistent memory for AI coding agents

A 80/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
No votes yet

01 / About

What Omega Memory is.

OMEGA is a local-first memory system for AI coding agents that persists decisions, lessons, error patterns, and preferences across sessions. It runs entirely on your machine as an MCP server backed by a single SQLite database, with embeddings computed on-device by an ONNX model, so it works with Claude Code, Claude Desktop, Cursor, Claw Code, Windsurf, Cline, Codex, and any other MCP client without an API key or a cloud service.

Memories are stored with a type (decision, lesson, error, summary) and retrieved through a pipeline of vector similarity via sqlite-vec (384-dimension bge-small-en-v1.5 embeddings), FTS5 full-text search, type-weighted scoring that doubles decisions and lessons, contextual re-ranking by tag, project, and content, and query-time deduplication. Hooks installed into the agent's settings capture lessons and decisions from prompts automatically and surface relevant memories after edits, reads, and shell commands; a session-start hook delivers a briefing of recent memories and profile.

The memory lifecycle deduplicates by SHA256 hash, embedding similarity at 0.85 or above, and per-type Jaccard overlap; similar content (55 to 95%) appends to an existing memory; session summaries expire after a day while lessons and preferences are permanent; and related edges link each new memory to its top three neighbours. Python 3.11 or later is required. The core package can also be used as a library without the MCP server, and a separate omega-pro package adds coordination, routing, entity, knowledge, and profile modules that register into the same server process.

Features

  • 25 memory tools: omegastore, omegaquery, omegawelcome, omegaprofile, edit and delete, omegasimilar, omegatimeline, omegatraverse over the relationship graph, and omegaweekly_digest
  • Compaction and consolidation: cluster and summarize related memories, prune stale entries, and clean edges
  • Checkpoints and reminders: save and resume task state across sessions, and set time-based reminders
  • Session hooks: SessionStart, Stop, UserPromptSubmit, and PostToolUse handlers dispatched through a daemon socket with fail-open semantics
  • CLI: omega setup, doctor, status, query, store, timeline, stats, consolidate, compact, backup, validate, and logs
  • Library API: store, query, and remember functions for scripts and CI without a server process
  • Local storage: ~/.omega/omega.db for memories, embeddings, and edges; about 31 MB RSS at startup and about 337 MB after the model loads
  • omega-pro modules: 29 coordination tools for file and branch locking, task queues, and agent messaging; LLM routing across five providers; entity registry; document ingestion; and AES-256 encrypted profile storage

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

03 / Build

Build with Omega Memory.

Browse the catalogue for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Omega Memory →

Browse the catalogue