Run OpenAI Agents SDK on Mingles Router
The SDK defaults to the Responses API and to sending traces to OpenAI. Neither is what you want against a third-party endpoint, and both are one line.
Pick your model and paste your key — every command and value on this page updates live.
Endpoint: https://router.mingles.ai/v1 · key stays in your browser ·
no key entered? commands show <your-key>.
Wire it up
1. Point a client at the endpoint
Build the model explicitly so the SDK uses chat completions rather than its default:
from agents import (
Agent, AsyncOpenAI, OpenAIChatCompletionsModel,
Runner, set_tracing_disabled,
)
set_tracing_disabled(disabled=True)
client = AsyncOpenAI(
base_url="https://router.mingles.ai/v1",
api_key="<your-key>",
)
model = OpenAIChatCompletionsModel(model="deepseek-ai/DeepSeek-V4-Flash-0731", openai_client=client)
agent = Agent(name="Assistant", instructions="Be concise.", model=model)
print(Runner.run_sync(agent, "Hello").final_output)set_tracing_disabled is not optional unless you also hold a platform.openai.com key — tracing uploads to OpenAI and will fail without one.
2. Or switch the default API globally
If you would rather not build the model object on every agent:
from agents import set_default_openai_client, set_default_openai_api
set_default_openai_client(client)
set_default_openai_api("chat_completions")The SDK defaults to the Responses API, which most providers — including us — do not implement. Without this line you get errors that look like auth failures but are not.
Exact values
| base_url | https://router.mingles.ai/v1 |
| api_key | <your-key> |
| model | deepseek-ai/DeepSeek-V4-Flash-0731 |
Switch model
Handoffs, guardrails and function tools all ride on chat completions once the default API is switched.
Before you file a bug: read the limits
Reasoning models spend the output budget on internal thinking, output is capped at 8192 tokens, there is no KV cache, and there are no built-in web tools. Most “it broke on Mingles Router” reports are one of these. See Model limits & behavior →
Frequently asked
I get an error that looks like a 404 or an unsupported endpoint. +
Almost always the Responses API default. Either construct OpenAIChatCompletionsModel explicitly or call set_default_openai_api("chat_completions").
Do I need an OpenAI account at all? +
Not for inference. You do for tracing — which is why the snippet disables it.