# LycheeMem — Lightweight Long-Term Agent Memory

> LycheeMem provides lightweight long-term memory for agents (SQLite + LanceDB) with reranking and runtime plugins. Verified 233★; setup ~10–20 minutes.

## Install

Save the content below to `.claude/skills/` or append to your `CLAUDE.md`:

## Quick Use

```bash
pip install "lycheemem[rerank]"  # or: pip install lycheemem
lycheemem-cli
# Configure providers in .env (litellm format), then integrate via the repo’s Claude/OpenClaw/Hermes plugins.
```

## Intro

LycheeMem provides lightweight long-term memory for agents (SQLite + LanceDB) with reranking and runtime plugins. Verified 233★; setup ~10–20 minutes.

**Best for:** Agent builders who want long-term recall without running heavy infra like Neo4j

**Works with:** Python 3.9+, litellm providers, LangGraph-style runtimes, plugins/MCP integration

**Setup time:** 10-20 minutes

### Key facts (verified)

- GitHub: 233 stars · 8 forks · pushed 2026-05-13.
- License: Apache-2.0 · owner avatar + repo URL verified via GitHub API.
- README-verified entrypoint: `pip install "lycheemem[rerank]" && lycheemem-cli`.

## Main

- Use it when you need recall across long projects: README describes structured consolidation + adaptive retrieval over SQLite + LanceDB.

- Start with the CLI server (`lycheemem-cli`) and a minimal .env, then add reranking once you’ve confirmed end-to-end recall works.

- Prefer runtime-native plugins (README links Claude Code/OpenClaw/Hermes install guides) to wire memory into your agent loop automatically.

### Source-backed notes

- README documents installation via `pip install lycheemem` and a recommended extra `pip install "lycheemem[rerank]"` for transformer reranking.
- README says semantic memory uses SQLite + LanceDB and explicitly notes “no Neo4j required”.
- README includes multiple runtime integrations (Claude Code plugin, OpenClaw plugin, Hermes plugin) and an MCP-compatible mode.

### FAQ

- **Do I need a vector database server?**: No — README uses embedded storage (SQLite + LanceDB) instead of running a separate DB service.
- **Is reranking required?**: No — README says the core works without the `rerank` extra; it improves evidence selection when installed.
- **How do I integrate with Claude Code?**: README links a Claude Code plugin install guide under `claude-plugin/`.

## Source & Thanks

> Source: https://github.com/LycheeMem/LycheeMem
> License: Apache-2.0
> GitHub stars: 233 · forks: 8

---

<!-- ZH -->

## Quick Use

```bash
pip install "lycheemem[rerank]"  # or: pip install lycheemem
lycheemem-cli
# Configure providers in .env (litellm format), then integrate via the repo’s Claude/OpenClaw/Hermes plugins.
```

## Intro

LycheeMem 为 agent 提供轻量长期记忆（SQLite + LanceDB），含 rerank 与多运行时插件，并提供 CLI 一键启动服务。已验证 233★；约 10–20 分钟可接入。

**Best for:** 想要长期记忆但不想上重型基础设施（如 Neo4j）的 agent 开发者

**Works with:** Python 3.9+、litellm 生态、LangGraph 类运行时，以及插件/MCP 集成

**Setup time:** 10-20 minutes

### Key facts (verified)

- GitHub：233 stars · 8 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 "lycheemem[rerank]" && lycheemem-cli`。

## Main

- 适合长项目的跨会话记忆：README 描述在 SQLite + LanceDB 上做结构化整理与自适应检索。

- 先用 `lycheemem-cli` + 最小 .env 跑通链路，再在需要时启用 rerank 以提升证据选择质量。

- 优先用运行时插件接入：README 链接 Claude Code/OpenClaw/Hermes 的安装指南，可把记忆自动接进 agent 循环。

### Source-backed notes

- README 提供 `pip install lycheemem`，并推荐 `pip install "lycheemem[rerank]"` 以启用 transformer rerank。
- README 说明语义记忆采用 SQLite + LanceDB，并明确标注“不需要 Neo4j”。
- README 给出多运行时集成：Claude Code/OpenClaw/Hermes 插件，以及 MCP 兼容模式。

### FAQ

- **需要单独跑向量数据库吗？**：不需要。README 使用嵌入式存储（SQLite + LanceDB），不必额外部署 DB 服务。
- **必须安装 rerank 吗？**：不必。README 表示核心系统可独立工作；安装 `rerank` 主要提升证据选择质量。
- **怎么接入 Claude Code？**：README 在 `claude-plugin/` 下提供 Claude Code 插件安装指南链接。

## Source & Thanks

> Source: https://github.com/LycheeMem/LycheeMem
> License: Apache-2.0
> GitHub stars: 233 · forks: 8


---
Source: https://tokrepo.com/en/workflows/lycheemem-lightweight-long-term-agent-memory
Author: Skill Factory