Start here

PI-Desktop is a local-first AI coding agent desktop client. The app keeps the workspace, host process, agent runtime, and provider configuration visible and inspectable while still making everyday coding work feel direct.

Choose a path

If you want to…	Start with
See what the app looks like	Screens
Run recurring local tasks	Scheduled tasks
Add an MCP catalog and its key	MCP market
Understand what is shipped	Product scope
Learn how the system fits together	Architecture
Trace a protocol or storage boundary	Runtime specs
Build an extension	Plugin development
Understand why a decision exists	ADR index
Validate a user-visible change	E2E test plan

The Chinese entry follows this same path: open the bilingual guide to read the localized orientation and jump into the matching topic map.

The mental model

Renderer UI  →  Electron orchestration  →  Rust host core
      ↓                    ↓                       ↓
  transcript          pi Node sidecar          SQLite + processes

The renderer owns presentation. Electron main coordinates desktop capabilities: window lifecycle, IPC routing, process supervision, the update client, and the plugin, MCP-bridge, and optional loopback MCP-control services. The Rust host owns tool execution and the workspace sandbox, the permission gateway, plugin host services, RPC, and persistence. The pi sidecar owns the agent loop and provider-facing model work.

Working with the docs

The documentation is English-first at the source-of-truth level. The 中文入口 provides the same information architecture and a complete translated companion for every specification. Each Chinese page links back to its English source; technical identifiers remain unchanged so search and cross-reference paths stay stable. Use global search when you know a term, protocol method, or decision number; use the sidebar when you are exploring a domain.

Before you change a boundary

1. Read the relevant spec.
2. Check the linked ADRs and the decisions log.
3. Update the E2E scenario when behavior is user-visible or protocol-visible.
4. Run the narrowest useful validation, then record the result in the change.

See the AI development workflow and change checklist for the complete repository rules.