Python SDK

The Python SDK is a fully async client for building agents on the Athyr platform. It provides the same orchestration patterns, middleware, and resilience features as the Go SDK, designed for Python’s async/await paradigm.

Installation

pip install git+https://github.com/athyr-tech/athyr-sdk-python.git

Requires Python 3.10+ and a running Athyr server.

Quick Start

import asyncio
from athyr import AthyrAgent, AgentCard, CompletionRequest, Message

async def main():
    async with AthyrAgent(
        "localhost:9090",
        agent_card=AgentCard(name="my-agent", description="My first Athyr agent"),
    ) as agent:
        resp = await agent.complete(CompletionRequest(
            model="llama3",
            messages=[Message(role="user", content="Hello!")]
        ))
        print(resp.content)

asyncio.run(main())

Features

Core Agent

• Connect/Disconnect — Lifecycle management with automatic reconnection
• Pub/Sub Messaging — Subscribe to subjects, publish messages, request/reply
• LLM Completions — Blocking and streaming completions via Athyr backends
• Memory Sessions — Conversation context with automatic summarization
• KV Storage — Key-value buckets for agent state
• Tool Calling — Decorator-based function calling support

Orchestration Patterns

Located in athyr.orchestration:

Middleware

• recover — Exception recovery
• timeout — Request timeouts
• retry — Automatic retries with backoff
• log_requests — Request/response logging
• Custom middleware via @middleware decorator

Tool Calling

Define tools with the @tool decorator:

from athyr import tool, ToolRegistry, run_tool_loop

@tool(description="Get weather for a city")
def get_weather(city: str, unit: str = "celsius") -> str:
    return f"Weather in {city}: sunny"

registry = ToolRegistry()
registry.register(get_weather)

response = await run_tool_loop(agent, request, registry)

Documentation

Full documentation and examples are maintained in the SDK repository:

• SDK README — Complete feature guide
• Examples — Working code samples

Available Examples

Next Steps

• Agents — Agent concepts and lifecycle
• LLM Gateway — LLM provider configuration
• State Management — Memory and KV details