Part 1 · Getting Started & Mental Model
This part helps you answer three questions quickly: what Agentao is, whether it fits your integration shape, and which path to start with.
Key terms in this Part
You only need these terms on the first pass. Full definitions live in Appendix G.
- Agentao — an embeddable Python agent runtime; call it directly from Python or drive it from another language through ACP · §1.1
- Tool / Skill / MCP — the common capability extensions: business functions, LLM-side instructions, external tool ecosystem · §1.2, Part 5
- Transport — the bridge between runtime and UI: streaming events, tool confirmation, user questions, max-iteration fallback · §1.2, Part 4
- Python SDK — in-process embedding for Python backends, data services, and batch jobs · §1.3, Part 2
- ACP — stdio JSON-RPC protocol embedding for IDE plugins, Node/Go/Rust hosts, and process-isolated deployments · §1.3, Part 3
Coverage
- 1.1 What is Agentao — product boundary, built-in capabilities, where it fits and where it does not
- 1.2 Core Concepts — Agent, Tool, Skill, Transport, Session, Working Directory
- 1.3 Integration Modes — Python SDK vs ACP, chosen by host language and isolation needs
- 1.4 Hello Agentao in 5 min — run a minimal working session first
- 1.5 Requirements — Python version, extras, credentials, OS, network, and disk layout
How to read
| Your situation | Recommended path |
|---|---|
| I just want to run it | 1.4 Hello → 1.2 Core Concepts |
| I need to decide whether it fits | 1.1 What is Agentao → 1.3 Integration Modes |
| My host is Python | 1.4 Hello → Part 2 |
| My host is not Python / I need process isolation | 1.3 Integration Modes → Part 3 |
| I am preparing for production | 1.5 Requirements → Part 6 |
Mental model
Agentao is not a chat UI and not a one-shot function call. It is the agent runtime inside your application: session state, tool calls, permissions, memory, event streams, and cross-language protocol. Your host owns the product experience, business policy, and deployment boundary.