finds.dev← search

// the find

volcengine/MineContext

★ 5,382 · Python · Apache-2.0 · updated May 2026

MineContext is your proactive context-aware AI partner(Context-Engineering+ChatGPT Pulse)

MineContext is a desktop app (Electron + Python backend) that takes periodic screenshots of your screen, runs them through a vision-language model, and builds a local context store you can query or get proactive summaries from. It's a self-hosted, local-first alternative to tools like Rewind or ChatGPT Pulse — built by ByteDance's volcengine team, so expect the default model recommendations to point at Doubao.

Local-first by default — SQLite + ChromaDB, all data stays on disk at a predictable path, no cloud sync required. The architecture is cleanly separated: Electron handles the UI and native capture, FastAPI handles the context pipeline, and the two talk over a local HTTP proxy — each layer is independently replaceable. OpenAI-compatible API protocol support means you can swap in any local model via LMStudio without changing application code. The context capture roadmap is unusually honest about what's done versus planned, and the P0/P1/P4/P5 priority tiers are realistic rather than vaporware.

The setup story is rough — you need uv, a Python venv, pnpm, and a backend build step before the Electron app even launches; first-run installs the Python environment at startup, which takes two minutes with no progress feedback. Screenshot-only context capture is a significant limitation right now: all the interesting sources (files, MCP, browser extension, mobile) are unimplemented and marked P1–P5, so what ships is basically a glorified screen recorder with LLM summarization. The strong recommendation to use Doubao models and the Lark/WeChat community channels make it clear this is primarily built for a Chinese developer audience — non-Chinese users will hit friction with model recommendations and community resources. No mention of how screen capture handles sensitive content like passwords or banking screens; the privacy story is 'local-first' but there's no apparent redaction or exclusion list.

View on GitHub →

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →