Agent Loop docs

Hello World across providers

Start with a scripted client to see the agent loop run without an API key. FakeClient returns the response you give it and records the request the loop sent.

from agentloop import Agent
from agentloop.provider import ModelResponse, StopReason, Usage
from agentloop.providers.fake import FakeClient
from agentloop.transcript import TextPart

client = FakeClient(
    ModelResponse(
        content=(TextPart(text="Hello, world!"),),
        stop_reason=StopReason.END_TURN,
        usage=Usage(input_tokens=3, output_tokens=4),
        model="fake-1",
        provider="fake",
    )
)
agent = Agent([], client=client)
print(agent.run("Say hello."))
print(f"model requests: {len(client.requests)}")

Output:

Hello, world!
model requests: 1

To call a live provider, replace the FakeClient and Agent construction above with Agent.from_env. The same run() call works for every built-in provider:

import os

from agentloop import Agent

with Agent.from_env(os.environ, tools=[]) as agent:
    print(agent.run("Say hello."))

Save the live example as hello.py, choose one configuration below, and run python hello.py. The reply comes from the selected model, so its exact text can vary. Agent.from_env receives os.environ explicitly; the library does not read the process environment or a .env file on its own.

Anthropic

pip install 'based-models-agentloop[anthropic]'
export LLM_PROVIDER=anthropic
export ANTHROPIC_API_KEY='your-key'
python hello.py

When ANTHROPIC_API_KEY is omitted, the Anthropic SDK may use another credential source it supports, such as an authenticated CLI profile. The default model is claude-opus-5; set ANTHROPIC_MODEL to choose another.

OpenAI

pip install 'based-models-agentloop[openai-compat]'
export LLM_PROVIDER=openai
export OPENAI_API_KEY='your-key'
python hello.py

OpenAI uses the Responses API and model gpt-5.1 by default. Set OPENAI_MODEL to choose another model or OPENAI_API=completions for Chat Completions compatibility.

OpenRouter

pip install 'based-models-agentloop[openai-compat]'
export LLM_PROVIDER=openrouter
export OPENROUTER_API_KEY='your-key'
python hello.py

The default model is openrouter/free; set OPENROUTER_MODEL to choose another.

pip install 'based-models-agentloop[openai-compat]'
export LLM_PROVIDER=modal
export MODAL_ENDPOINT_URL='https://your-deployment.modal.run'
python hello.py

The default endpoint model is Qwen/Qwen3.6-35B-A3B-FP8; set MODAL_ENDPOINT_MODEL to match your deployment. For an authenticated Modal proxy, set both MODAL_PROXY_TOKEN_ID and MODAL_PROXY_TOKEN_SECRET. Setting only one is a configuration error.

Keep credentials out of source control. See configuration for the full list of settings.