weaiw / trove-ai

What is trove-ai?

Trove AI is a self-hostable read-later app with built-in AI processing, designed specifically for Chinese internet content. You paste a WeChat article URL or forward it via the WeChat bot, and it automatically extracts the full content, generates a summary, tags it, creates vector embeddings for semantic search, and builds a knowledge graph connecting related pieces in your library. The TypeScript frontend (Next.js) talks to a Python FastAPI backend backed by PostgreSQL with pgvector for vector storage, Redis for caching, and supports any OpenAI-compatible LLM provider. You run it with one Docker compose command, own your data, and can sync to Obsidian as a backup.

Why is it gaining traction?

Pocket shut down in 2024. Omnivore shut down too. Users who spent years curating personal knowledge libraries lost everything overnight. Trove AI solves this with true data sovereignty: you host it yourself, your data stays on your server, and the companion Obsidian plugin means your knowledge survives even if the project dies. The other hook is deep Chinese platform integration. Unlike generic read-later tools that struggle with WeChat's dynamic rendering or Xiaohongshu's anti-bot measures, Trove has built-in parsers for eight-plus Chinese platforms and falls back to headless Chromium rendering when needed. The auto knowledge graph is also unique -- every article you save finds its three semantically closest siblings, so your library self-organizes over time.

Who should use this?

Product managers and researchers who accumulate hundreds of saved articles but never revisit them. Engineers who want their weekly reading to compound into structured knowledge rather than disappearing into a browser tab graveyard. Privacy-conscious users unwilling to trust a startup with their reading habits. Anyone burned by Pocket or Omnivore shutdowns and now demanding data sovereignty. The WeChat bot is especially useful for people who do most of their reading on mobile -- you forward a link from any chat, and it's in your library with a summary within seconds.

Verdict

With a credibility score of 0.8500000238418579%, Trove AI is a well-architected, thoughtfully designed project that fills a real gap -- self-hosted, AI-powered, Chinese-internet-native read-later with Obsidian backup. The code quality looks solid (async FastAPI, proper data isolation, clean router/service separation), Docker deployment works out of the box, and the feature set is more complete than any competitor. At 16 stars, it's early-stage and relatively unknown, which means the community is small and bugs may surface as you push the edges. If you want the full feature set (WeChat bot, knowledge graph, learning paths) and self-hosting is acceptable, it's worth the risk. If you need stability and a large community, wait until it gains more traction.