How dicklesworthstone/pi_agent_rust works
pi_agent_rust is a native terminal AI coding agent. The main workflow starts with a user prompt or resumed session, configures a model and credentials, runs an agent turn that can call coding tools, then streams and persists the result. The product also documents extension security, MCP servers, multi-agent work, planning, LSP, browser/computer control, and multiple interaction modes. Relationships below use README statements and SOURCE INDEX references; unsampled subsystems are represented only where their paths and capabilities are documented.
Main entry points: agent.rs Agent loop · app.rs Application entry · ext_session.rs Interactive mode · provider.rs Provider interface
Components 23
User interaction
- Application entry
src/app.rs - Interactive mode
src/interactive/ext_session.rs - RPC mode
legacy_pi_mono_code/pi-mono/packages/coding-agent/src/modes/rpc/rpc-mode.ts
Agent runtime
- Agent loop
src/agent.rs - Capability context
src/agent_cx.rs - Context compaction
src/compaction_worker.rs - Plan sessions
src/plan/session.rs - Subagent execution
src/subagents/execution.rs
Models and providers
- Configuration
src/config.rs - Authentication
src/auth.rs - Model registry
src/models.rs - Provider interface
src/provider.rs - Provider backends
src/providers/mod.rs - Streaming HTTP
src/http/client.rs
Tools and integrations
- Coding tools
src/tools.rs - Extensions
src/extensions/extension_manager_impl.rs - MCP servers
src/mcp/manager.rs - Language server
src/lsp/client.rs - Browser control
src/browser/cdp.rs
Sessions and state
- Session persistence
src/session.rs - Agent memory
src/memory/shared.rs
Other components
- CLI user
- Model service
Connections 24
- CLI user → Application entry: prompts
README.md - Application entry → Configuration: loads settings
src/app.rs - Application entry → Model registry: selects models
src/app.rs - Application entry → Authentication: uses credentials
src/app.rs - Application entry → Agent loop: starts agent
- Application entry → Interactive mode: offers interaction
- Interactive mode → Session persistence: resumes sessions
src/interactive/ext_session.rs - Agent loop → Capability context: uses capabilities
src/agent_cx.rs - Agent loop → Provider interface: requests completions
src/provider.rs - Provider interface → Provider backends: implemented by
src/providers/mod.rs - Provider backends → Streaming HTTP: streams requests
src/providers/mod.rs - Streaming HTTP → Model service: sends requests
- Agent loop → Coding tools: invokes tools
- Agent loop → Session persistence: persists messages
- Agent loop → Context compaction: compacts context
- Agent loop → Plan sessions: supports planning
README.md - Agent loop → Subagent execution: runs subagents
README.md - Agent loop → Extensions: loads extensions
README.md - Extensions → Coding tools: gates tool calls
README.md - Agent loop → MCP servers: uses MCP tools
README.md - Agent loop → Language server: supports code intelligence
- Agent loop → Browser control: supports browser control
- Agent loop → Agent memory: uses shared memory
- Application entry → RPC mode: offers RPC mode