Chat & Orchestration

Io is your primary point of contact. Every message you send — whether via the web dashboard or Telegram — goes through Io's orchestrator.

How It Works

1. Message received → Router classifies complexity tier
2. Model selected → Fast, standard, or premium based on classification
3. System prompt built → Includes wiki context, squad roster, active skills
4. Decision made → Answer directly OR delegate to a squad

Direct Mode

When no squad exists for the relevant repository (or the question isn't project-specific), Io answers directly with its full tool set:

• File reading/editing
• Running commands
• Git operations
• Code search
• Wiki management
• Scheduling

Delegation Mode

When a squad exists for the repository in question, Io delegates automatically:

You: "Add dark mode to the settings page in my-app"
Io:  "I'll delegate this to the my-app squad..."
     → Creates objective → Squad executes → Results in inbox

Rolling Window Reset

To maintain context quality over long conversations, Io uses a rolling window:

1. After N messages (default: 50), the session resets
2. Before reset, the conversation is summarized
3. Summary is saved as a wiki episode (conversations/YYYY-MM-DD.md)
4. A fresh session starts, seeded with the summary + relevant wiki context

This ensures Io never loses important context while keeping the active window focused.

Streaming

Responses stream in real-time via WebSocket (web) or Telegram message editing. You see tokens as they're generated.