labazhou2024

labazhou2024 / memexa

Public

Self-hosted Chinese personal memory graph. Six sources, two LLMs, one graph.

github.comlabazhou2024memex bge-m3 chinese-nlp cli knowledge-graph llm-pipeline
16
1
89% credibility
Found May 19, 2026 at 17 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Memexa (镜我, meaning 'Mirror Self') is a self-hosted personal memory system that turns scattered Chinese data — from WeChat, QQ, email, browser history, AI conversations, and voice recordings — into a searchable knowledge graph. It runs entirely on your own computer, using AI to extract structured information from raw messages and store it with citations back to the original text. You can query your memory using natural questions like 'Who is Alice?' or 'What was the whole story behind X?' and get organized answers. The system is designed to work alongside AI coding assistants, letting them answer questions about your personal life using real context from your own conversations.

How It Works

1
💬 You hear about a tool that remembers everything

A friend mentions memexa — a way to search through years of WeChat chats, emails, and voice memos all at once.

2
🚀 You try it instantly with no signup

Running one simple command shows you a demo with fake conversations — you see exactly how it works in 30 seconds.

3
🔑 You connect your AI thinking assistant

You enter your AI service details so memexa can understand and organize your conversations intelligently.

4
You choose where your memories live
📧
Email inbox

Connect your email account to remember every promise, deadline, and conversation

💬
WeChat messages

Export your chat history and search through years of group conversations

🎙️
Voice memos

Upload audio files and have them transcribed and searchable

5
✨ Your memories become searchable

Behind the scenes, memexa reads your messages, understands context, and stores everything in a private database on your computer.

6
🔍 You ask anything — and get answers with proof

Ask 'What did my professor want in the last email?' or 'Who was Alice talking to about the project?' and get answers with links back to the original messages.

🎉 Your personal memory assistant is ready

Everything stays on your computer, private and fast. Your AI helper can now answer questions about your life using real context.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 17 to 16 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is memexa?

Memexa is a self-hosted memory graph for personal data, built specifically for Chinese content. It pulls from six sources—WeChat, QQ, 飞书, 钉钉, email, and audio recordings—and uses a two-LLM pipeline to extract structured facts while keeping the original text verbatim. When you query it, you get cards with citations back to the source sentence. The CLI exposes 14 subcommands for different question types: who is someone, what happened over time, what a project looks like across sources, pending commitments, and synthesized answers via LLM reflection. Everything runs locally with PostgreSQL and pgvector under the hood. Python-based, ships with Docker compose for the backend, and works as a subprocess for AI agents like Claude Code or Cursor today, with MCP support planned.

Why is it gaining traction?

The hook is Chinese-native data handling at a personal scale. Most memory graph tools target English use cases or generic documents. Memexa specifically handles Chinese IM platforms, email, and voice recordings with speaker diarization and voice embedding. The two-LLM approach—gate plus extractor—keeps costs low (DeepSeek runs about ¥0.30 per 1000 messages) while maintaining citation quality. The demo runs in 30 seconds with no backend or API key required, giving developers an instant taste of what the system does.

Who should use this?

Developers building AI agents that need to answer questions about a user's personal context. Researchers studying Chinese conversational patterns who want structured extraction with provenance. Power users who live in WeChat or QQ and want a local searchable memory layer. Anyone running self-hosted infrastructure who prefers their data stays local rather than hitting third-party APIs.

Verdict

At 16 stars and v0.1.1, this is early-stage software with a credibility score of 0.9%. The documentation is thorough and the architecture is thoughtful, but the low star count and alpha status mean you should budget time for breaking changes. If you need Chinese personal memory management and don't mind an early-stage project, it's worth a look—start with `memexa demo` to see if the query patterns fit your needs.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.