MineContext

MineContext is an Apache-2.0 desktop application from ByteDance's VolcEngine team that continuously captures screenshots and digital activity, runs them through a vision-language pipeline, and stores the result as queryable personal context. Instead of waiting for prompts, it proactively delivers daily and weekly summaries, surfaces todos, and generates tips based on what it has observed you doing.

It is aimed at developers and power users who want a local 'second brain' without sending data to a SaaS vendor. You bring your own API keys, defaulting to ByteDance's Doubao models (Doubao-Seed-1.6-flash for vision, Doubao-embedding-vision for retrieval) but it accepts any OpenAI-compatible endpoint, including local models. A FastAPI server with WebSocket support exposes a REST API for building custom clients on top of the captured context store.

The project is early and clearly a strategic showcase for the Doubao/VolcEngine stack, so the smoothest path uses ByteDance models; mileage with local Llama or GPT will depend on the OpenAI-compat layer. Privacy-wise the capture loop is local, but anything you send to a hosted model leaves the machine on its terms, not yours.