AI agent application platform · desktop, web and Mini Program
A platform for using and collaborating with AI agents through desktop, web and Mini Program clients, with conversations, skills, tools, model access and remote execution. Build frontend interfaces and desktop integrations, and contribute to backend services and multi-device sessions with AI assistance.
- Session ownership across devices
- Desktop update recovery and device information
Context
The platform gives users desktop and browser access to AI agents. The interface must represent session ownership, execution state, skill selection and failures alongside streamed messages.
My contribution
- Build and iterate on React conversation interfaces, including message streaming, skill suggestions and remote session access.
- Integrate Tauri device information and desktop capabilities; handle remaining Windows processes, update failures and page recovery.
- Contribute to FastAPI services, database migrations and active-session coordination with AI-assisted development.
A representative problem: one session, multiple devices
A local “generating” flag cannot tell whether another device is already running the same conversation.
I contributed to server-side active-session leases and the corresponding ownership, device information and message-state UI, with targeted tests for the service and interface behavior.
Engineering approach
Keep desktop-specific behavior, session APIs and message rendering separate. Investigate both UI state and server request lifecycles when diagnosing failures, and review and test AI-assisted backend changes.
Related articles
- Technical notes: Stream parsing: one network read is not one message
- Technical notes: One conversation, several devices: making execution ownership explicit
- React
- TypeScript
- React Router
- Zustand
- Ant Design
- Tailwind CSS
- Tauri
- Rust
- Taro
- SSE
- MCP
- FastAPI
- PostgreSQL