Capability Map
This map covers the repository's public module families and links to practical guides. It distinguishes runtime mechanisms from application responsibilities.
| Capability | Guide | Host still owns |
|---|---|---|
| Agent loop and execution entry points | Execution | Run lifetime and user isolation |
| Configuration groups and options | Configuration | Model-specific budgets |
| Tools, hooks, concurrency, dual output | Tools | Authorization and side effects |
| Files, shell, Git, web, browser, terminals | Built-in catalog | Sandbox and external dependencies |
| Extensions and lifecycle composition | Extensions | Resource ownership and ordering |
| System prompts, rules, profiles, caching | Prompts | Trust and provider caching policy |
| Skills and progressive disclosure | Skills | Trusted packages and administration |
| Messages, images, thinking, history conversion | Messages | Data policy and provider compatibility |
| Typed JSON output | Structured output | Semantic validation |
| Provider adapters | Providers | Credentials and model entitlement |
| Retry, failover, selection, artifacts | Reliability | Budgets, retention, safe replay |
| Context engines and summaries | Compaction | Useful budgets and information loss |
| Events and OpenTelemetry | Observability | Exporters, redaction, reconciliation |
| Four-mode execution and streams | Runnables | Correct aggregation/cancellation |
| DAG, Pregel, typed graphs, graph-as-tool | Graphs | Merge semantics and termination |
| Pipelines, routing, handoffs, task tools | Workflows | Specialist boundaries and cleanup |
| Snapshots, JSONL trees, branches | Sessions | Durability, permissions, retention |
| Checkpoints, interruption, restoration | Recovery | Human approval and exactly-once policy |
| HTTP/SSE and thread operations | HTTP | Authentication and deployment |
| MCP tools, resources, prompts | MCP | Trusted server and transport policy |
| A2A tasks and remote agents | A2A | Remote auth and compatibility |
| ACP editor sessions | ACP | Factory and capability implementation |
| AG-UI execution events | AG-UI | Web renderer and state reconciliation |
| A2UI declarative surfaces | A2UI | Catalog-aware client renderer |
| Terminal UI and event adapter | TUI | Terminal capabilities and run lifetime |
| RAG Loader/Embedder/Indexer/Retriever | Retrieval | Concrete backends and document policy |
| Templates, file queue, fuzzy matching, util | Utilities | Validation and mutation safety |
Reading Paths
- First agent: Quick Start, Configuration, Tools, Providers.
- Interactive application: Execution, Messages, Observability, HTTP or TUI.
- Multi-step application: Runnables, Graphs, Workflows, Recovery.
- Production deployment: Reliability, Sessions, Production Checklist, Testing.
The package index remains the symbol-level entry point. This site follows main, and the map is not a claim that every external service or model supports every listed feature.