Aion Python SDK

A Python library that serves agents over the Agent-to-Agent (A2A) protocol and runs them on the Aion platform. Bring a LangGraph or Google ADK agent, point an aion.yaml at it, and run aion serve — you get an A2A endpoint with an agent card, streaming and conversation state, plus a terminal client to try it out.

✨ Features

• Framework agnostic: LangGraph and Google ADK agents get the same configuration and the same A2A surface.
• A2A protocol compliant: agent card, streaming and conversation state without writing protocol code.
• Four lines of configuration: an agent needs one field — where to find it. Everything else has a default.
• No account required: runs fully locally; add credentials to reach the Aion model service and MCP tools.
• Multi-agent: a built-in proxy fronts several agents behind one endpoint, and aion chat talks to any of them from the terminal.

🚀 Getting started

Prerequisites

• Python 3.12, 3.13 or 3.14
• Node.js 22+ — only for the bundled aion chat client

🔧 Installation

Install the SDK plus the extra for the framework you use.

The *-authoring extras are the toolkit on its own — Aion models, tools and messaging inside your agent — for agents that something other than aion serve runs.

💡 Example

A LangGraph agent, served over A2A, with four lines of configuration.

1. Write the agent. A plain LangGraph graph, returned uncompiled: Aion compiles it and attaches the checkpointer, so conversation state comes for free.

   # agent.py
   from typing import Annotated, TypedDict
   
   from langchain_core.messages import AIMessage, AnyMessage
   from langgraph.graph import END, START, StateGraph
   from langgraph.graph.message import add_messages
   
   
   class State(TypedDict):
       messages: Annotated[list[AnyMessage], add_messages]
   
   
   def reply(state: State) -> dict:
       question = state["messages"][-1].content
       return {"messages": [AIMessage(content=f"Echo: {question}")]}
   
   
   def create_graph() -> StateGraph:
       workflow = StateGraph(State)
       workflow.add_node("reply", reply)
       workflow.add_edge(START, "reply")
       workflow.add_edge("reply", END)
       return workflow
   

2. Point Aion at it. path is the only field an agent needs.

   # aion.yaml
   aion:
     agents:
       support:
         path: "./agent.py:create_graph"
   

3. Serve it. The port is assigned automatically and printed on startup.

   aion serve
   

4. Talk to it from another terminal.

   aion chat
   

That is a complete local run — no account, no credentials. To connect the agent to the platform, add AION_CLIENT_ID and AION_CLIENT_SECRET to a .env beside the configuration.

For Google ADK the configuration is the same — see the ADK quickstart.

📚 Documentation

Full documentation lives at docs.aion.to:

• LangGraph quickstart · Google ADK quickstart
• aion.yaml reference · Environment variables
• CLI reference · Multiple agents and the proxy
• Troubleshooting

🤝 Contributing

Contributions are welcome. For development setup and how to run the test suites, see the Development Guide.

📄 License

This project is licensed under the MIT License. See the LICENSE file for more details.