Agno | Doubleword Inference API

Agno (formerly Phidata) provides an OpenAILike model class designed for custom OpenAI-compatible endpoints.

Install

pip install agno openai

Configure

from agno.agent import Agent
from agno.models.openai import OpenAILike

agent = Agent(
    model=OpenAILike(
        id="{{selectedModel.id}}",
        base_url="https://api.doubleword.ai/v1",
        api_key="{{apiKey}}",
    )
)

agent.print_response("Say hello.")

Prompt caching

Pass the system prompt as a Message whose content block carries cache_control:

from agno.agent import Agent
from agno.models.message import Message
from agno.models.openai import OpenAILike

agent = Agent(
    model=OpenAILike(
        id="{{selectedModel.id}}",
        base_url="https://api.doubleword.ai/v1",
        api_key="{{apiKey}}",
    ),
    system_message=Message(
        role="system",
        content=[{
            "type": "text",
            "text": "…large, stable instructions (~1024-token floor)…",
            "cache_control": {"type": "ephemeral", "ttl": "1h"},
        }],
    ),
)

response = agent.run("What is 2 + 2?")
print(response.metrics.cache_read_tokens)

Agno forwards the block unchanged. It reports cache reads but not writes, so compare cache_read_tokens across two runs. See the prompt caching guide.

Batch pricing with Autobatcher

For background tasks where latency is not critical, use Autobatcher to transparently route requests through the Batch API at reduced cost:

pip install agno openai autobatcher

from autobatcher import BatchOpenAI
from agno.agent import Agent
from agno.models.openai import OpenAILike
import asyncio

client = BatchOpenAI(
    api_key="{{apiKey}}",
    base_url="https://api.doubleword.ai/v1",
)

agent = Agent(
    model=OpenAILike(
        id="{{selectedModel.id}}",
        async_client=client,
    )
)

async def main():
    response = await agent.arun("Say hello.")
    print(response.content)
    await client.close()

asyncio.run(main())

BatchOpenAI is a drop-in AsyncOpenAI subclass that collects requests and submits them as batch jobs automatically, cutting inference costs by up to 90%.