Introduction

What is AgentRoute?

AgentRoute is a Python SDK for building, running, and shipping AI agents. You write agent logic with plain Python functions, run it against any OpenAI-compatible model, and layer in tools, persistent memory, conversation history compaction, and structured output as your agent grows.

The API is designed to be progressive — the simplest program is three lines, and every capability is opt-in from there.

hello.py

from agentroute import Agent
 
agent = Agent("my-bot", model="claude-sonnet-4")
print(agent.run("Tell me a joke"))

Set AGENTROUTE_API_KEY (or OPENROUTER_API_KEY) and the same code works for claude-sonnet-4, gpt-4o, gemini-2.0-flash, deepseek-v3, and more — one key, every model.

The layered API

Layer 0 — talk to a model. Construct an Agent, call agent.run(...).
Layer 1 — add tools. Decorate a function with @agent.tool and the model can call it.
Memory & history. Persist conversation and facts, and compact long transcripts to fit the context window.
Structured output. Ask for a typed Pydantic model back instead of text, with validators that can request a retry.

Start with the Quickstart, then dig into the Concepts and the SDK reference.