Skip to main content
The OpenAI Agents SDK builds agentic loops — model calls, tool calls, handoffs — on top of an OpenAI client. Because that client speaks the OpenAI Chat Completions protocol, you can point it at any OpenAI-compatible endpoint. Set the base URL to Valar and the SDK runs your agents on Valar models with no other changes.

​
Run an agent on Valar

1

Install the SDK

pip install openai-agents
2

Set your API key

Export your Valar key so the client can read it from the environment.
export VALAR_API_KEY="your-api-key"
3

Point the client at Valar

Create an AsyncOpenAI client with Valar’s base URL, then wrap it in an OpenAIChatCompletionsModel and hand that to your agent. The model id is a Valar model — see Models for the full list.
from agents import Agent, Runner, OpenAIChatCompletionsModel
from openai import AsyncOpenAI
import os

client = AsyncOpenAI(base_url="https://api.valarhq.ai/v1", api_key=os.environ["VALAR_API_KEY"])
agent = Agent(
    name="Assistant",
    instructions="You are a concise assistant.",
    model=OpenAIChatCompletionsModel(model="moonshotai/Kimi-K2.6", openai_client=client),
)
result = Runner.run_sync(agent, "Summarize the benefits of async inference in two sentences.")
print(result.final_output)
4

Add a tool

Decorate a Python function with @function_tool and attach it to the agent through tools. The SDK exposes the function’s signature and docstring to the model, runs the call when the model asks for it, and feeds the result back into the loop.
from agents import Agent, Runner, OpenAIChatCompletionsModel, function_tool
from openai import AsyncOpenAI
import os

client = AsyncOpenAI(base_url="https://api.valarhq.ai/v1", api_key=os.environ["VALAR_API_KEY"])

@function_tool
def get_weather(city: str) -> str:
    """Return the current weather for a city."""
    return f"The weather in {city} is sunny."

agent = Agent(
    name="Assistant",
    instructions="Use the tools you have to answer the question.",
    model=OpenAIChatCompletionsModel(model="moonshotai/Kimi-K2.6", openai_client=client),
    tools=[get_weather],
)
result = Runner.run_sync(agent, "What is the weather in Amsterdam?")
print(result.final_output)
The SDK’s hosted tracing sends trajectories to OpenAI and expects an OpenAI key. When you run purely on Valar, disable it with set_tracing_disabled(True) (imported from agents).
from agents import set_tracing_disabled

set_tracing_disabled(True)
Valar offers two completion windows: Now (asap) for interactive work and Standard (standard) for cheaper, latency-tolerant work. See Completion windows and Inference modes for how to choose between them.