II-Agent is an open-source AI agent framework designed to build and deploy agents for real work. The project has completed its beta phase and is licensed under Apache-2.0, 100% open source. Whether you are an independent developer, a research team, or an enterprise building internal tools, you can run, replicate, and extend it, with no black box and no vendor lock-in. Users can bring their own API keys (BYOK) and have full control over costs and model providers.
Main capabilities include:
- Mobile app development: generate a complete mobile app from a short prompt.
- Website app development: generate a complete website app from a short prompt.
- Storybook generation: create a fully illustrated picture book from a single prompt.
- Video and image generation: support multiple models in one workflow.
- Real-time editing: real-time editing of websites, slides, and storybooks.
- Plan mode: visualize project planning before building.
- Fast research and deep research: provide quick answers or multi-step investigations.
- Interactive website generation: turn a research brief into a complete website containing structure, visuals, citations, and embedded Q&A.
Automation and integration:
- Built-in and custom skills: reuse workflows and connect GitHub-based processes.
- App integrations: support Gmail, Slack, GitHub, Notion, Google Calendar, Discord, Dropbox, Canva, and more.
- Faster execution speed: significantly improved compared with the early beta version.
How to use: Requires installing Docker, uv (command: curl -LsSf https://astral.sh/uv/install.sh | sh), Node.js and npm (or use nvm). After running, the backend is located at http://localhost:8000，前端位于 http://localhost:1420，PostgreSQL at localhost:5432, Redis at localhost:6379, MinIO (S3-compatible storage) at http://localhost:9001（用户名/密码：minioadmin/minioadmin）。
Configuration: II-Agent supports multiple LLM providers. It can be configured in model_configs.yaml, or via MODEL_CONFIGS in .env. For complete configuration options (including Vertex AI, Azure, and self-hosted models), see model_configs.example.yaml.
How to run: You can use Make commands, or run the full stack via Docker Compose (no local Python/Node required). For more details, refer to the official guide.
Use cases: Suitable for developers, research teams, and enterprises that need to quickly build apps, generate content, conduct research, or integrate multiple tools.
Limitations: The materials do not provide specific system requirements, performance metrics, pricing, or fee information, nor do they explain the details of the supported model list.
