MineContext Documentation

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

5,313 stars Python Python 1 file · ~9,646 tokens #agent#context-engineering#electron#embedding-models#javascript#memory#proactive-ai#python
RAW Doc

File: README.md

Table of Contents

🔗 Related Project: Check out OpenViking - An open-source Context Database designed for AI Agents. OpenViking unifies Memories, Resources, and Skills through a "file system paradigm", providing the infrastructure layer for sophisticated context management.

👋🏻 What is MineContext

MineContext is a proactive context-aware AI partner. By utilizing screenshots and content comprehension (with future support for multi-source multimodal information including documents, images, videos, code, and external application data), it can see and understand the user's digital world context. Based on an underlying contextual engineering framework, it actively delivers high-quality information such as insights, daily/weekly summaries, to-do lists, and activity records.

🚀 Key Features

MineContext focuses on four key features: effortless collection, intelligent resurfacing, proactive delivery, and a context engineering architecture.

  1. 📥 Effortless Collection
    Capable of gathering and processing massive amounts of context. Designed storage management enables extensive collection without adding mental burden.
  2. 🚀 Proactive Delivery
    Delivers key information and insights proactively in daily use. It extracts summarized content from your context—such as daily/weekly summaries, tips, and todos—and pushes them directly to your homepage.
  3. 💡 Intelligent Resurfacing
    Surfaces relevant and useful context intelligently during creation. Ensures assisted creativity without overwhelming you with information.
  4. 🎯 Context Engineering Architecture
    Supports the complete lifecycle of multimodal, multi-source data—from capture, processing, and storage to management, retrieval, and consumption—enabling the generation of six types of intelligent context.

🔏 Privacy Protection

Local-First

MineContext places a high priority on user privacy. By default, all data is stored locally in the following path to ensure your privacy and security.

text
~/Library/Application Support/MineContext/Data

Local AI model

In addition, we support custom model services based on the OpenAI API protocol. You can use fully local models in MineContext, ensuring that any data does not leave your local environment.

🏁 Quick Start

1. Installation

Click Github Latest Release to Download

Note: Starting from v0.1.5, MineContext supports Apple notarization, so you no longer need to disable the quarantine attribute. If you're using an older version, please refer to the previous documentation for instructions.

2. Enter Your API Key

After the application launches, please follow the prompts to enter your API key. (Note: On the first run, the application needs to install the backend environment, which may take about two minutes).

We currently support services from Doubao, OpenAI, and custom models. This includes any local models or third-party model services that are compatible with the OpenAI API format.

We recommend using LMStudio to run local models. It provides a simple interface and powerful features to help you quickly deploy and manage them.

Considering both cost and performance, we recommend using the Doubao model. The Doubao API Key can be generated in the API Management Interface.

After obtaining the Doubao API Key, you need to activate two models in the Model Activation Management Interface: the Visual Language Model and the Embedding Model.

  • Visual Language Model: Doubao-Seed-1.6-flash

  • Embedding Model: Doubao-embedding-vision

The following is the filling process after obtaining the API Key:

3. Start Recording

Enter [Screen Monitor] to enable the system permissions for screen sharing. After completing the setup, you need to restart the application for the changes to take effect.

After restarting the application, please first set your screen sharing area in [Settings], then click [Start Recording] to begin taking screenshots.

4. Forget it

After starting the recording, your context will gradually be collected. It will take some time to generate value. So, forget about it and focus on other tasks with peace of mind. MineContext will generate to-dos, prompts, summaries, and activities for you in the background. Of course, you can also engage in proactive Q&A through [Chat with AI].

5. Backend Debugging

MineContext supports backend debugging, which can be accessed at http://localhost:1733.

1.View Token Consumption and Usage

2.Configure Interval for Automated Tasks

3.Adjust System Prompt for Automated Tasks

🎃 Contribution Guide

🎨 Frontend Architecture

The MineContext frontend is a cross-platform desktop application built with Electron, React, and TypeScript, providing a modular, maintainable, and high-performance foundation for desktop development.

Core Tech Stack

Technology Description
Electron Allows for the development of cross-platform desktop applications using web technologies.
React A component-based UI library for building dynamic user interfaces.
TypeScript Provides static type checking to enhance code maintainability.
Vite A modern frontend build tool optimized for Electron.
Tailwind CSS A utility-first CSS framework for rapid and consistent UI styling.
pnpm A fast and efficient package manager suitable for monorepo projects.

Core Architecture

The project follows a standard Electron architectural design, clearly separating the code for the main process, preload scripts, and renderer process to ensure security and maintainability.

text
frontend/
├── src/
│ ├── main/     # Electron main process (window management, lifecycle, IPC)
│ ├── preload/  # Preload script, securely bridging Node APIs and the renderer process
│ └── renderer/ # React frontend interface (renderer process)
│
├── packages/
│ └── shared/   # Common utilities, IPC channels, logging, and constant definitions
│
├── build/      # Build resources (icons, platform configurations)
├── dist/       # Build artifacts generated by electron-builder
├── externals/  # External dependencies (Python scripts, binaries, etc.)
├── resources/  # Static assets (icons, templates, images)
└── scripts/    # Development and build helper scripts

  1. Main Process (src/main/) is responsible for:

    • Managing application windows
    • Handling lifecycle events (startup, quit, activate)
    • Establishing secure IPC communication
    • Integrating with backend services (Python and system APIs)

  2. Preload Script (src/preload/) is responsible for:

    • Securely exposing Node.js APIs to the renderer process
    • Handling IPC communication with the main process
    • Implementing cross-process resource access

  3. Renderer Process (src/renderer/) is responsible for:

    • Implementing the user interface with React
    • Managing global state with Jotai and Redux
    • Utilizing an efficient styling system based on Tailwind CSS
    • Implementing dynamic loading and performance optimization mechanisms

  4. Build and Packaging are responsible for:

    • electron-vite.config.ts — Configures the build logic for both the main and renderer processes (aliases, plugins, etc.).
    • electron-builder.yml — Defines packaging and distribution configurations for Windows, macOS, and Linux.

💻 Frontend Usage

Build Backend

Before starting frontend development, you need to build the backend first:

bash
uv sync
source .venv/bin/activate
./build.sh

Install Dependencies

Due to package version issues, using a domestic PyPI mirror is not currently supported. Please run the following command to ensure you are using the original PyPI source:

bash
pip config unset global.index-url
cd frontend
pnpm install

Development and Debugging

During local development, it is normal for the screen capture area selection to be slow. Please wait, as this issue does not exist in the packaged application.

bash
pnpm dev

Application Packaging

To build APP for macOS:

bash
pnpm build:mac
# Data Path
# ～/Library/Application\ Support/MineContext

The executable files generated by the packaging process will be stored in the MineContext/frontend/dist directory.

🏗️ Backend Architecture

MineContext adopts a modular, layered architecture design with clear separation of concerns and well-defined responsibilities for each component.

Core Architecture Components

text
opencontext/
├── server/             # Web server and API layer
├── managers/           # Business logic managers
├── context_capture/    # Context acquisition modules
├── context_processing/ # Context processing pipeline
├── context_consumption/# Context consumption and generation
├── storage/            # Multi-backend storage layer
├── llm/               # LLM integration layer
├── tools/             # Tool system
└── monitoring/        # System monitoring

Layer Responsibilities

  1. Server Layer (server/)

    • FastAPI-based RESTful API
    • WebSocket support for real-time communication
    • Static file serving and template rendering

  2. Manager Layer (managers/)

    • CaptureManager: Manages all context capture sources
    • ProcessorManager: Coordinates context processing pipeline
    • ConsumptionManager: Handles context consumption and generation
    • EventManager: Event-driven system coordination

  3. Context Capture Layer (context_capture/)

    • Screenshot monitoring
    • Document monitoring
    • Extensible capture interface for future sources

  4. Processing Layer (context_processing/)

    • Document chunking strategies
    • Entity extraction and normalization
    • Context merging and deduplication
    • Multi-modal content processing (text, images)

  5. Storage Layer (storage/)

    • Multi-backend support (SQLite, ChromaDB)
    • Vector storage for similarity search
    • Unified storage interface

  6. LLM Integration (llm/)

    • Support for multiple LLM providers (OpenAI, Doubao)
    • VLM (Vision-Language Model) integration
    • Embedding generation services

🚀 Backend Usage

Installation

We recommend using uv for fast and reliable package management:

bash
# Clone repository
git clone https://github.com/volcengine/MineContext.git
cd MineContext

# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Sync dependencies (automatically creates virtual environment)
uv sync

Configuration

  1. Basic Configuration (config/config.yaml):
yaml
server:
  host: 127.0.0.1
  port: 8765
  debug: false

embedding_model:
  provider: doubao # options: openai, doubao
  api_key: your-api-key
  model: doubao-embedding-vision-250615

vlm_model:
  provider: doubao # options: openai, doubao
  api_key: your-api-key
  model: doubao-seed-1-6-flash-250828

capture:
  enabled: true
  screenshot:
    enabled: true # enable screenshot capture
    capture_interval: 5 # capture interval in seconds
  1. Prompt Templates (config/prompts_*.yaml):
    • prompts_en.yaml: English prompt templates
    • prompts_zh.yaml: Chinese prompt templates

Running the Server

bash
# Start with default configuration
uv run opencontext start

# Start with custom config
uv run opencontext start --config /path/to/config.yaml

# Start with custom port (useful for avoiding conflicts)
uv run opencontext start --port 1733

Available Options:

  • --config: Path to configuration file
  • --host: Host address (default: from config or localhost)
  • --port: Port number (default: from config or 1733)

Priority: Command-line arguments > Config file > Default values

Alternatively, you can activate the virtual environment manually:

bash
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
pip install -e .
opencontext start --port 1733

💎 The Philosophy Behind the Name

The naming of MineContext also reflects the team's ingenuity. It signifies both "my context" and "mining context." It draws inspiration from the core philosophy of Minecraft—openness, creativity, and exploration.

If vast amounts of context are like scattered "blocks," then MineContext provides a "world" where you can freely build, combine, and create. Users can reimagine and create new content based on the collected massive context and generate high-quality information.

🎯 Target User

Target User Category Specific Roles/Identities Core Needs/Pain Points
Knowledge Workers Researchers, Analysts Navigating vast amounts of information, improving information processing and analysis efficiency
Content Creators Writers, Bloggers Craving endless inspiration, optimizing content creation workflows
Lifelong Learners Students, Researchers Building systematic knowledge systems, efficiently managing and connecting learning materials
Project Managers Product Managers, Project Managers Integrating multi-source information and data, ensuring project alignment and decision-making efficiency

🔌 Context-Source

We will prioritize the expansion of Context Sources according to the following plan, and we warmly welcome everyone to actively contribute code to our efforts.

  • P0: Digital life and public information loop (PC screen capture and link upload)
  • P1: Personal text context loop (file upload, file tracking)
  • P2: AI and common office context loop (MCP, meeting notes)
  • P3: High-quality information acquisition loop (DeepResearch and RSS)
  • P4: Personal deep context loop (WeChat, QQ chat data acquisition, mobile screenshots)
  • P5: Physical world context loop (smart wearable synchronization, smart glasses synchronization)
Context Capture Capability Context Source Priority Completion Status
Screen Screenshot User PC Information P0 ✅
Note Editing Application Internal Creation Information P0 ✅
Link Upload Internet Information P0
File Upload Structured Documents P1
File Upload Unstructured Documents P1
File Upload Images P1
File Upload Audio P4
File Upload Video P4
File Upload Code P4
Browser Extension AI Conversation Records P2
Browser Extension Refined Internet Information P5
Meeting Records Meeting Information P2
RSS Consultation Information P3
Deep Research High-Quality Research Analysis P3
Application MCP/API Payment Records P4
Application MCP/API Research Papers P3
Application MCP/API News P4
Application MCP/API Emails P4
Application MCP/API Notion P2
Application MCP/API Obsidian P2
Application MCP/API Slack P4
Application MCP/API Jira P4
Application MCP/API Figma P2
Application MCP/API Linear P4
Application MCP/API Todoist P4
Memory Bank Migration Import User Memory P4
WeChat Data Capture WeChat Chat History P4
QQ Data Capture QQ Chat History P4
Mobile Screenshot Monitor User Mobile End Information P4
Smart Glasses Data Sync Physical World Interaction Records P5
Smart Bracelet Data Sync Physiological Data P5

🆚 Comparison with Familiar Application

MineContext vs ChatGPT Pulse

  • 🖥️ Comprehensive Digital Context:
    MineContext captures your entire digital workflow by reading from screen screenshots, providing a rich, visual context of your daily activities and applications. ChatGPT Pulse, in contrast, is limited to the context of a single text-based conversation.
  • 🔒 Local-First Data & Privacy:
    Your data is processed and stored entirely on your local device, ensuring complete privacy and security without relying on cloud servers. ChatGPT Pulse requires data to be sent to and stored on OpenAI's servers.
  • 🚀 Proactive & Diverse Insights:
    MineContext delivers a wider variety of intelligent, auto-generated content—including daily summaries, actionable todos, and activity reports—not just simple tips. ChatGPT Pulse primarily offers reactive assistance within the chat interface.
  • 🔧 Open Source & Customizable:
    As an open-source project, MineContext allows developers to freely inspect, modify, and build upon the codebase for complete customization. ChatGPT Pulse is a closed, proprietary product with no option for modification.
  • 💰 Cost-Effective API Usage:
    MineContext avoids the need for a costly $200/month Pro subscription by allowing you to use your own API key, giving you full control over your spending. ChatGPT Pulse's advanced features are locked behind its expensive premium tier.

MineContext vs Dayflow

  • 💡 Richer, Proactive Insights:
    MineContext delivers a more diverse range of automated, intelligent content—including concise summaries, actionable todos, and contextual tips—going beyond basic activity tracking. DayFlow primarily focuses on logging user activity.
  • 🧠 Context-Aware Q&A & Creation:
    MineContext enables you to ask questions and generate new content based on your captured context, unlocking wider application scenarios like content drafting and project planning. DayFlow is limited to passive activity recording and review.
  • ✨ Superior Activity Generation & Experience:
    MineContext produces activity records with greater clarity and detail, featuring a more intuitive and interactive dashboard for a seamless user experience. DayFlow's activity logs are more basic with limited interactivity.

👥 Community

Community and Support

  • GitHub Issues: Errors and issues encountered while using MineContext.
  • Email Support: Feedback and questions about using MineContext.
  • WeChat Group: Discuss SwanLab usage and share the latest AI technologies.

Star History

📃 License

This repository is licensed under the Apache 2.0 License.

File: frontend/externals/python/window_inspector/README.md

macOS Window Inspector

Steps

  1. Install dependencies
bash
python3 -m venv venv
source venv/bin/activate
pip3 install -r requirements.txt
pip3 install pyinstaller
pyinstaller --onedir --name window_inspector window_inspector.py (optional)
pyinstaller window_inspector.spec
  1. Test run
bash
./dist/window_inspector/window_inspector

  1. Calling in Electron

    Add the following to build.extraResources in package.json:

json
{
  "build": {
    "extraResources": [
      {
        "from": "python/window_inspector",
        "to": "bin/window_inspector"
      }
    ]
  }
}

File: frontend/externals/python/window_capture/README.md

macOS Window Capture

Steps

  1. Install dependencies
bash
python3 -m venv venv
source venv/bin/activate
pip3 install -r requirements.txt
pip3 install pyinstaller
pyinstaller --onedir --name window_capture window_capture.py (optional)
pyinstaller window_capture.spec
  1. Test run
bash
./dist/window_capture/window_capture

  1. Calling in Electron

    Add the following to build.extraResources in package.json:

json
{
  "build": {
    "extraResources": [
      {
        "from": "python/window_capture",
        "to": "bin/window_capture"
      }
    ]
  }
}

File: frontend/README.md

MineContext

Project Setup

Build Backend

for macos

bash
uv sync
source .venv/bin/activate
./build.sh

for windows

not support yet

Install

bash
cd frontend
pnpm install

Development

bash
pnpm dev

Build APP

bash
# For macOS
pnpm build:mac
# Data Path
# ～/Library/Application\ Support/MineContext

Data Path

～/Library/Application\ Support/MineContext

File: README_zh.md

目录

🔗 相关项目：欢迎了解 OpenViking - 一个专为 AI Agents 设计的开源上下文数据库。OpenViking 通过"文件系统范式"统一管理记忆、资源和技能三类上下文，为复杂的上下文管理提供基础设施层。

👋🏻 MineContext 是什么

MineContext 是一个具有上下文感知能力的主动式 AI 伙伴。它基于屏幕截图+内容理解的方式（未来还将支持其他来源的多模态信息，包括文档、图片、视频、代码、外部应用数据），能够看到并看懂用户的数字世界上下文，然后再基于底层的上下文工程框架，主动推送洞察、日/周总结 、待办、活动记录等高质量信息，同时支持用户基于 Context 和生成的信息进行再创作。

🚀 核心功能

MineContext 专注于四个核心功能：无负担收集、主动推送、智能浮现和上下文工程架构。

  1. 📥 无负担收集
    支持收集和处理海量的 Context，并通过设计存储管理来实现海量收集却没有心智负担。
  2. 🚀 主动推送
    支持日常主动推送关键信息和洞见，能够提炼 Context 中的总结信息，比如每日总结，每周总结，tips，todo，主动推送到主页。
  3. 💡 智能浮现（实现中）
    支持创作时智能浮现，可以随时浮现相关有用的 Context，确保辅助创作又不会被淹没
  4. 🎯 上下文工程架构
    支持多模态、多源数据的完整生命周期——从捕获、处理和存储到管理、检索和消费——支持生成六种类型的智能上下文。

🔏 隐私保护

本地存储

MineContext 非常注重用户隐私，所有数据都默认保存在本地如下路径，确保您的隐私和安全。

text
~/Library/Application Support/MineContext/Data

本地模型

此外我们支持了 OpenAI API 协议的自定义模型服务，您可以在 MineContext 中使用全本地模型，做到任何数据不上云。

🏁 快速开始

1. 安装

点击 Github Latest Release 下载最新版本。

注意：从 v0.1.5 版本开始，MineContext 已支持 Apple 公证，因此不再需要禁用隔离属性。如果您使用的是旧版本，请参考之前的文档获取相关说明。

2. 输入您的 API 密钥

应用程序启动后（首次运行时需要安装后端环境，约需等待两分钟），请根据引导输入您的 API 密钥。目前我们支持豆包、OpenAI 以及自定义模型服务，包括任何兼容 OpenAI API 格式的本地模型或第三方模型服务。
我们推荐使用 LMStudio 来运行本地模型，它提供了简单的界面和强大的功能，能够帮助您快速部署和管理本地模型。

综合成本和性能，我们推荐使用豆包模型，豆包模型的 API-Key 可以在 API 管理界面 生成。

获取豆包 API 之后需要在 模型开通管理界面 开通视觉语言模型和向量化两个模型。

  • 视觉语言模型：Doubao-Seed-1.6-flash

  • 向量化模型：Doubao-embedding-vision

以下是获取了 API Key 后的填写流程：

3. 开始记录

进入【Screen Monitor】启用屏幕分享的系统权限，设置完之后需要重新启动应用使其生效。

重新启动应用后，请先在【Settings】设置您的屏幕共享区域，然后点击【Start Recording】开始截图。

4. 忘掉它

启动记录后，您的上下文将逐渐被收集。这会需要一些时间才能产生价值。所以说，忘记它，安心专注于其他任务吧。MineContext 将会在后台为您生成待办事项、提示、摘要和活动。当然，您也可以通过【Chat with AI】进行主动问答。

5. 后台调试

MineContext 支持在http://localhost:1733 进行后台调试。

1.支持查看 Token 用量与使用情况

2.支持主动推送任务的时间间隔设置

3.支持调整主动推送任务的系统提示词

🎃 贡献指南

🎨 前端架构

MineContext 前端 是一个基于 Electron、React 和 TypeScript 构建的 跨平台桌面应用程序，提供模块化、可维护且高性能的桌面开发基础。

核心技术栈

技术 描述
Electron 允许使用 Web 技术开发跨平台桌面应用程序。
React 用于构建动态用户界面的基于组件的 UI 库。
TypeScript 提供静态类型检查，增强代码可维护性。
Vite 针对 Electron 优化的现代前端构建工具。
Tailwind CSS 用于快速且一致地设计 UI 的实用优先 CSS 框架。
pnpm 适用于 monorepo 项目的快速高效的包管理器。

核心架构

该项目遵循 标准的 Electron 架构设计，将主进程、预加载脚本和渲染进程的代码清晰分离，以保证安全性和可维护性。

text
frontend/
├── src/
│ ├── main/ # Electron 主进程（窗口管理、生命周期、IPC 通信）
│ ├── preload/ # 预加载脚本，安全桥接 Node API 与渲染进程
│ └── renderer/ # React 前端界面（渲染进程）
│
├── packages/
│ └── shared/ # 通用工具、IPC 通道、日志与常量定义
│
├── build/ # 构建资源（图标、平台配置）
├── dist/ # 由 electron-builder 生成的构建产物
├── externals/ # 外部依赖（Python 脚本、二进制文件等）
├── resources/ # 静态资源（图标、模板、图片）
└── scripts/ # 开发与构建辅助脚本

1、主进程 (src/main/)负责：

  • 管理应用窗口
  • 处理生命周期事件（启动、退出、激活）
  • 建立安全的 IPC 通信
  • 与后端服务（Python 与系统 API）集成

2、预加载脚本 (src/preload/)负责：

  • 安全地将 Node.js API 暴露给渲染进程
  • 处理与主进程的 IPC 通信
  • 实现跨进程的资源访问

3、渲染进程 (src/renderer/)负责：

  • 实现基于 React 的用户界面
  • 使用 Jotai 与 Redux 管理全局状态
  • 基于 Tailwind CSS 的高效样式体系
  • 动态加载与性能优化机制

4、构建与打包负责：

  • electron-vite.config.ts — 同时配置主进程与渲染进程的构建逻辑（别名、插件等）。
  • electron-builder.yml — 定义针对 Windows、macOS、Linux 的打包与分发配置。

💻 前端使用

构建后端

在开始前端开发之前，需要先构建后端：

bash
uv sync
source .venv/bin/activate
./build.sh

安装依赖

由于包版本原因，目前不支持使用国内 PYPI 源，请输入以下命令，确保使用的是原始 PYPI 环境

bash
pip config unset global.index-url
cd frontend
pnpm install

开发调试

本地开发时，截屏范围获取较慢属于正常现象，等待即可，打包应用无此问题。

bash
pnpm dev

应用打包

为 macOS 平台构建应用：

bash
pnpm build:mac

打包生成的可执行文件会存放在MineContext/frontend/dist目录下。

🏗️ 后端架构

MineContext 采用模块化、分层的架构设计，各组件职责明确，关注点分离。

核心架构组件

text
opencontext/
├── server/             # Web服务器和API层
├── managers/           # 业务逻辑管理器
├── context_capture/    # 上下文获取模块
├── context_processing/ # 上下文处理流水线
├── context_consumption/# 上下文消费和生成
├── storage/            # 多后端存储层
├── llm/               # LLM集成层
├── tools/             # 工具系统
└── monitoring/        # 系统监控

各层职责

  1. 服务器层 (server/)

    • 基于 FastAPI 的 RESTful API
    • 支持实时通信的 WebSocket
    • 静态文件服务和模板渲染

  2. 管理器层 (managers/)

    • CaptureManager：管理所有上下文捕获源
    • ProcessorManager：协调上下文处理流水线
    • ConsumptionManager：处理上下文消费和生成
    • EventManager：事件驱动的系统协调

  3. 上下文捕获层 (context_capture/)

    • 屏幕截图监控
    • 文档监控
    • 可扩展的捕获接口，支持未来的来源

  4. 处理层 (context_processing/)

    • 文档分块策略
    • 实体提取和规范化
    • 上下文合并和去重
    • 多模态内容处理（文本、图像）

  5. 存储层 (storage/)

    • 多后端支持（SQLite、ChromaDB）
    • 用于相似性搜索的向量存储
    • 统一的存储接口

  6. LLM 集成 (llm/)

    • 支持多个 LLM 提供商（OpenAI、豆包）
    • VLM（视觉-语言模型）集成
    • 嵌入生成服务

🚀 后端使用

安装

我们推荐使用 uv 进行快速、可靠的包管理：

bash
# 克隆仓库
git clone https://github.com/volcengine/MineContext.git
cd MineContext

# 安装 uv（如果尚未安装）
curl -LsSf https://astral.sh/uv/install.sh | sh

# 同步依赖（自动创建虚拟环境）
uv sync

配置

  1. 基本配置 (config/config.yaml)：
yaml
server:
  host: 127.0.0.1
  port: 8765
  debug: false

embedding_model:
  provider: doubao # 选项：openai, doubao
  api_key: your-api-key
  model: doubao-embedding-vision-250615

vlm_model:
  provider: doubao # 选项：openai, doubao
  api_key: your-api-key
  model: doubao-seed-1-6-flash-250828

capture:
  enabled: true
  screenshot:
    enabled: true # 开启截图捕获
    capture_interval: 5 # 截图间隔（秒）
  1. 提示模板 (config/prompts_*.yaml)：
    • prompts_en.yaml：英文提示模板
    • prompts_zh.yaml：中文提示模板

运行服务器

bash
# 使用默认配置启动
uv run opencontext start

# 使用自定义配置启动
uv run opencontext start --config /path/to/config.yaml

# 使用自定义端口启动
uv run opencontext start --port 1733

可用选项：

  • --config：配置文件路径
  • --host：主机地址（默认：配置文件中的值或 localhost）
  • --port：端口号（默认：配置文件中的值或 1733）

优先级：命令行参数 > 配置文件 > 默认值

或者，你也可以手动激活虚拟环境：

bash
source .venv/bin/activate  # Windows系统：.venv\Scripts\activate
pip install -e .
opencontext start --port 1733

💎 MineContext 与我的世界

MineContext 的命名，也体现了团队的巧思。既是“我的上下文”，更要“挖掘上下文”。它借鉴了 MineCraft（我的世界）的核心理念——开放、创造与探索。

如果说海量的 Context 是散落各处的“方块”，那么 MineContext 提供的就是一个让你能够自由搭建、组合、创造的“世界”。用户除了接收到主动推送的信息外，还能够基于收集到的海量 Context 和生成的高质量信息进行再创作。

🎯 目标用户

目标用户类别 具体角色/身份 核心需求/痛点
知识工作者 研究人员、分析师 浏览海量信息，提高信息处理和分析效率
内容创作者 作家、博主 渴求无尽灵感，优化内容创作工作流程
终身学习者 学生、研究者 建立系统化知识体系，高效管理和连接学习材料
项目经理 产品经理、项目经理 整合多源信息和数据，确保项目一致性和决策效率

🔌 上下文来源

我们将按照以下计划优先扩展上下文来源，热烈欢迎大家积极贡献代码。

  • P0：数字生活和公共信息循环（PC 屏幕捕获和链接上传）
  • P1：个人文本上下文循环（文件上传、文件跟踪）
  • P2：AI 和常见办公上下文循环（MCP、会议记录）
  • P3：高质量信息获取循环（DeepResearch 和 RSS）
  • P4：个人深度上下文循环（微信、QQ 聊天数据获取、手机截图）
  • P5：物理世界上下文循环（智能穿戴同步、智能眼镜同步）
上下文捕获能力 上下文来源 优先级 完成状态
屏幕截图 用户 PC 信息 P0 ✅
笔记编辑 应用内创作信息 P0 ✅
链接上传 互联网信息 P0
文件上传 结构化文档 P1
文件上传 非结构化文档 P1
文件上传 图像 P1
文件上传 音频 P4
文件上传 视频 P4
文件上传 代码 P4
浏览器扩展 AI 对话记录 P2
浏览器扩展 提炼的互联网信息 P5
会议记录 会议信息 P2
RSS 咨询信息 P3
Deep Research 高质量研究分析 P3
应用 MCP/API 支付记录 P4
应用 MCP/API 研究论文 P3
应用 MCP/API 新闻 P4
应用 MCP/API 电子邮件 P4
应用 MCP/API Notion P2
应用 MCP/API Obsidian P2
应用 MCP/API Slack P4
应用 MCP/API Jira P4
应用 MCP/API Figma P2
应用 MCP/API Linear P4
应用 MCP/API Todoist P4
记忆库迁移导入 用户记忆 P4
微信数据捕获 微信聊天历史 P4
QQ 数据捕获 QQ 聊天历史 P4
手机截图监控 用户移动端信息 P4
智能眼镜数据同步 物理世界交互记录 P5
智能手环数据同步 生理数据 P5

🆚 与同类应用的比较

MineContext vs ChatGPT Pulse

  • 🖥️ 全面的数字世界上下文：
    MineContext 通过读取屏幕截图捕获您的整个数字工作流程，提供丰富的、可视化的日常活动和应用程序上下文。相比之下，ChatGPT Pulse 仅限于单个基于文本的对话上下文。
  • 🔒 本地优先数据与隐私：
    您的数据完全在本地设备上处理和存储，确保完全的隐私和安全，无需依赖云服务器。ChatGPT Pulse 要求数据发送到并存储在 OpenAI 的服务器上。
  • 🚀 更加多样化的主动推送：
    MineContext 提供更广泛的智能自动生成内容——包括每日摘要、可操作的待办事项和活动报告——而不仅仅是简单的提示。ChatGPT Pulse 仅在每天早上提供 5-10 个提示。
  • 🔧 开源可定制：
    作为一个开源项目，MineContext 允许开发人员自由检查、修改和构建代码库，实现完全定制。ChatGPT Pulse 是一个封闭的专有产品，无法修改。
  • 💰 经济实惠的 API 使用：
    MineContext 通过允许您使用自己的 API 密钥，避免了每月 200 美元的昂贵 Pro 订阅费用，让您完全控制支出。ChatGPT Pulse 的高级功能被锁定在其昂贵的高级订阅后面。

MineContext vs Dayflow

  • 💡 更丰富、更主动的洞察：
    MineContext 提供更多样化的自动智能内容——包括简明摘要、可操作的待办事项和上下文提示——超越基本的活动跟踪。DayFlow 仅记录用户活动。
  • 🧠 上下文感知的问答与创作：
    MineContext 允许您基于捕获的上下文提问和生成新内容，解锁更广泛的应用场景，如内容起草和项目规划。DayFlow 仅限于被动的活动记录和回顾。
  • ✨ 更优质的活动生成与体验：
    MineContext 生成的活动记录更加清晰和详细，具有更直观和交互式的仪表板，提供无缝的用户体验。DayFlow 的活动日志更基本，交互性有限。

👥 社区

社区与支持

  • GitHub Issues：使用 MineContext 时遇到的错误和问题。
  • 邮件支持：关于使用 MineContext 的反馈和问题。
  • 微信群：讨论 MineContext 使用并分享最新 AI 技术。

Star History

📃 许可证

本仓库在 Apache 2.0 许可证下发布。

Explore Related Technologies & Topics

View All 9,111+ Repositories →

Discover similar high-velocity repositories, agent skills, and OpenAPI specifications across the ecosystem.

#agent Browse agent catalog → #context-engineering Browse context-engineering catalog → #electron Browse electron catalog → #embedding-models Browse embedding-models catalog → #javascript Browse javascript catalog → #memory Browse memory catalog → ⚡ MCP Servers Model Context Protocol → 🧠 Agent Skills Prompt & rule files → 🔌 OpenAPI Specs REST & Swagger schemas →