# omega-memory — Persistent Memory for Coding Agents

> OMEGA is a local-first, cross-model memory system for coding agents. It runs an MCP server plus hooks so recall is automatic across sessions.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

```bash
pip install omega-memory[server]
omega setup
omega doctor
```

## Intro

OMEGA is a local-first, cross-model memory system for coding agents. It runs an MCP server plus hooks so recall is automatic across sessions.

**Best for:** Developers who want cross-session memory without cloud lock-in

**Works with:** Python 3.11+, MCP clients, Claude Code/Cursor/Codex (via setup)

**Setup time:** 4-10 minutes

### Key facts (verified)

- GitHub: 135 stars · 22 forks · pushed 2026-05-13.
- License: Apache-2.0 · owner avatar + repo URL verified via GitHub API.
- README-verified entrypoint: `pip install omega-memory[server] # Full install (memory + MCP server)`.

## Main

- Use the server+hooks path if you want ambient memory: setup wires MCP + session hooks so recall happens at session start, not only when you remember to query.

- Start by storing explicit decisions (architecture, style, constraints), then rely on semantic query when a new session begins to keep continuity.

- If you only need a library (no server), use the README’s core-only install — but expect fewer integrations than the MCP server flow.

### Source-backed notes

- README Quick Install shows `pip install omega-memory[server]` followed by `omega setup` and `omega doctor`.
- README states `omega setup` downloads an ONNX embedding model (~90MB) and registers an MCP server for supported clients.
- README lists a memory-tool table (25 tools) and documents a CLI section for setup/doctor/status.

### FAQ

- **Does it send data to the cloud?**: README frames it as local-first; any network traffic comes from your LLM provider, not the memory store itself.
- **Can I use it without MCP?**: Yes — README shows a library-only install, but integrations are strongest with the MCP server flow.
- **What’s a good first memory?**: Store one durable rule (e.g., linting or architecture constraint) and verify it’s recalled on the next session start.

## Source & Thanks

> Source: https://github.com/omega-memory/omega-memory
> License: Apache-2.0
> GitHub stars: 135 · forks: 22

---

<!-- ZH -->

## Quick Use

```bash
pip install omega-memory[server]
omega setup
omega doctor
```

## Intro

OMEGA 是本地优先的跨模型记忆系统：提供 MCP server + hooks，让编码 agent 跨会话自动回忆决策与偏好。`pip install omega-memory[server]` 后跑 `omega setup/doctor`，嵌入模型本地下载（约 90MB）。

**Best for:** 想要跨会话记忆、又不想把数据交给云端/单一厂商的开发者

**Works with:** Python 3.11+；MCP 客户端；Claude Code/Cursor/Codex（setup 自动接入）

**Setup time:** 4-10 minutes

### Key facts (verified)

- GitHub：135 stars · 22 forks；最近更新 2026-05-13。
- 许可证：Apache-2.0；作者头像与仓库链接均已通过 GitHub API 复核。
- README 中核对过的入口命令：`pip install omega-memory[server] # Full install (memory + MCP server)`。

## Main

- 如果你想要“自动记忆”，走 server+hooks 路线：setup 会把 MCP 与会话 hooks 接好，让回忆在 SessionStart 自动发生。

- 先显式存关键决策（架构/风格/约束），再在新会话开始时用语义检索保持连续性，减少重复解释。

- 只想当库用（不跑 server）也可以按 README 的 core-only 安装，但集成能力会比 MCP server 少。

### Source-backed notes

- README 的 Quick Install 给出 `pip install omega-memory[server]`，并随后执行 `omega setup` / `omega doctor`。
- README 说明 `omega setup` 会下载 ONNX 嵌入模型（约 90MB）并为对应客户端注册 MCP server。
- README 列出 25 个 memory tools，并在 CLI 章节写明 setup/doctor/status 等命令。

### FAQ

- **会把数据上传云端吗？**：README 把它定位为 local-first；网络流量主要来自你使用的 LLM provider，而不是记忆库本身。
- **不接 MCP 也能用吗？**：能。README 有 library-only 安装，但最完整的集成在 MCP server + hooks 流程。
- **第一条记忆存什么最值？**：先存一条长期规则（如 lint/架构约束），然后新会话开始时验证是否被自动召回。

## Source & Thanks

> Source: https://github.com/omega-memory/omega-memory
> License: Apache-2.0
> GitHub stars: 135 · forks: 22


---
Source: https://tokrepo.com/en/workflows/omega-memory-persistent-memory-for-coding-agents
Author: MCP Hub