What is MineContext?
MineContext is a proactive context-aware AI partner: a desktop application that captures screenshots of your digital world, understands their content, and builds a picture of what you have been working on. It sits in the Notes & Knowledge Bases category of the AI Productivity & Writing tools and is open-source, written in Python and licensed Apache-2.0.
Overview
MineContext is a proactive context-aware AI partner: a desktop application that captures screenshots of your digital world, understands their content, and builds a picture of what you have been working on. Instead of waiting for you to ask a question, it delivers insights, daily and weekly summaries, to-do lists and activity records on its own.
The project organises itself around four ideas — effortless collection, intelligent resurfacing, proactive delivery, and a context-engineering architecture underneath them. Screenshots and content comprehension are the current capture path, with the roadmap pointing at multi-source multimodal input including documents, images, videos, code and data from external applications.
It is local-first by design and ships as a native app for macOS and Windows, released from the project's own GitHub releases page (Apple notarization has been supported since v0.1.5). Model choice is yours: it works with Doubao, OpenAI, and any local or third-party service that speaks the OpenAI-compatible API format, and the README points at LMStudio for running models locally. Two models are needed — a vision-language model to read the captures and an embedding model to index them.
MineContext is published by Volcengine, ByteDance's cloud arm, under the Apache 2.0 licence. The same team maintains OpenViking, a related open-source context database for AI agents that unifies memories, resources and skills under a file-system paradigm.
What it does
- Continuous screenshot capture with content comprehension, no manual note-taking
- Proactive delivery of insights, daily and weekly summaries, to-dos and activity records
- Local-first architecture — captures and the context store stay on your machine
- Works with Doubao, OpenAI, or any OpenAI-compatible endpoint, including local models via LMStudio
- Native desktop builds for macOS and Windows, with a separately runnable Python backend for debugging
- Context-engineering architecture underneath, shared with the OpenViking context database
Getting started
MineContext is distributed as a packaged desktop app. Installation is a download; the work is in pointing it at a model provider.
Download the app
Grab the latest macOS .dmg or Windows installer from the project's GitHub releases page at github.com/volcengine/MineContext/releases. From v0.1.5 onward the macOS build is Apple-notarized, so no quarantine workaround is needed.
Enter an API key on first launch
The app prompts for a key when it starts. On the very first run it also installs its backend environment, which the README says takes about two minutes.
Activate two models
MineContext needs a vision-language model to read your captures and an embedding model to index them. With Doubao, the README's recommended pairing on cost and performance, activate Doubao-Seed-1.6-flash and Doubao-embedding-vision in the Volcengine Ark model management console. Any OpenAI-compatible endpoint, including a local one served by LMStudio, works instead.
Start recording, then leave it alone
Turn on capture and go back to work. The value shows up later, when the app resurfaces what it saw as summaries, to-dos and activity records.
Commands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Reconstruct what you actually worked on this week without having kept notes
- Get a to-do list assembled from what you were doing rather than from what you remembered to write down
- Keep an AI assistant aware of your work context while keeping the captures on your own machine
- Run the whole loop against a local model when screen contents cannot leave the device
How MineContext compares
MineContext alongside other open-source notes & knowledge bases tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| SiYuan | ★ 47k | A self-hosted, block-level knowledge base with Markdown WYSIWYG editing, SQL queries over blocks, spaced-repetition flashcards and AI writing and Q&A through an OpenAI-compatible API. |
| notebooklm-py | ★ 20.1k | An unofficial Python API, CLI and MCP server for Google's Gemini Notebook (formerly NotebookLM): bulk source import, cited Q&A, and batch export of audio, video, slides, quizzes, mind maps and data tables. |
| MineContext | ★ 5.5k | A local-first desktop app that watches your screen and resurfaces what matters |
| Smart Connections | ★ 5.5k | Obsidian plugin that surfaces semantically related notes and excerpts as you write, using a bundled local embedding model that needs no API key and works offline and on mobile. |
| OpenKnowledge | ★ 4.8k | A local-first WYSIWYG markdown IDE and LLM wiki that keeps a knowledge base as plain files in git and exposes it to Claude, Codex and OpenCode over MCP. |
| ReadAny | ★ 2.9k | A local-first e-book reader for desktop and mobile that indexes your library into a local vector store, then answers questions about a book with hybrid vector + BM25 retrieval, alongside highlights, notes, TTS and WebDAV sync. |
| LLM Wiki Compiler | ★ 2.2k | CLI, MCP server and TypeScript SDK (llmwiki) that compiles raw sources into an interlinked, citation-traceable markdown wiki for people and agents. |
| LLM Wiki | ★ 1.7k | An open-source take on Karpathy's LLM Wiki that has Claude compile your documents, web clips and highlights into a cited, cross-linked Markdown wiki over MCP, with a web app, graph viewer and Chrome clipper. |