Skip to content

Blueprint Overview

The blueprint is organized around the major systems inside an agent CLI, but the current teaching path is intentionally Python-first.

  1. Tool system
  2. Tool execution pipeline
  3. Query engine and conversation loop
  4. Bootstrap and startup
  5. Permission system
  6. Session and state management
  7. Multi-agent orchestration
  8. MCP integration
  9. Slash command system
  10. Terminal UI
  11. IDE bridge protocol
  12. Memory and cost tracking

Python-first route

Reading tracks

Core runtime

Expansion layers

Advanced Patterns

  • Streaming Formats -- Wire formats and streaming protocols for real-time token delivery
  • YOLO Classifier -- Automatic tool-approval classification and trust scoring
  • Model Pricing -- Token cost tracking, model selection, and budget-aware routing
  • Hidden Modes -- Undocumented operational modes and internal feature flags

Operator experience

Chapter navigation

Use this section when

  • you want the full architecture in one place
  • you need a chapter-by-chapter reference
  • you are mapping blueprint ideas to your own runtime

The main document still lives at Full Blueprint, but the chapter pages are now the recommended reading path.

How to use this section

  • Use the chapter pages when you want Python-first implementation guidance.
  • Use the full blueprint when you want the exact long-form walkthrough and source-level provenance.
  • Use the build-better pages when you want to convert observations into design rules for your own harness.