volcengine/MineContext

Captures continuous screenshots and processes them through vision-language models to build a multimodal context database, then surfaces relevant information proactively via daily summaries and intelligent resurfacing during work. Built with a local-first architecture that stores all data on-device by default, while supporting both cloud APIs (OpenAI, Doubao) and fully local models compatible with OpenAI's protocol. Frontend uses Electron/TypeScript with a backend service layer that handles screenshot processing, embedding generation, and context retrieval for the retrieval-augmented workflow.

5,049 stars. Actively maintained with 3 commits in the last 30 days.