Use Outrun with LiteLLM.

Before you start

Have your API key and some Outrun credit ready. Keep your key private.

Setup Steps

1. 1
   Prepare your LiteLLM installation

   Use Python 3.10 or newer with a supported LiteLLM installation. The SDK examples need the litellm package; the Proxy section also requires its proxy extra. Select a supported version before installing or upgrading.

2. 2
   Keep your key in a local variable

   Provide OUTRUN_API_KEY only to your local process. Use your own key from Outrun → API keys or @outrunAPI_Bot → My API key. Never put it in source control, browser code or a shared configuration.

3. 3
   Use the OpenAI-compatible route

   Use a model string with the openai/ prefix, and keep api_base exactly https://api.outrun.to/v1. The prefix selects LiteLLM’s OpenAI-compatible Chat Completions route; it does not change which Outrun model ID is called.

4. 4
   Run one small request

   Run the SDK example below once, or start the loopback-only proxy and send its example request. Requests use Outrun credit. Check Activity before repeating an interrupted request.

Call Outrun from Python

Keep the openai/ prefix and replace the model ID with any available chat model from the model catalogue. Image, video and audio operations use their own endpoints, not this chat example.

import os
from litellm import completion

response = completion(
    model="openai/claude-sonnet-5-5",
    api_base="https://api.outrun.to/v1",
    api_key=os.environ["OUTRUN_API_KEY"],
    messages=[{"role": "user", "content": "Reply with OK only."}],
)
print(response.choices[0].message.content)

Stream the response

Add stream=True and consume the returned chunks. Use streaming only when the selected model’s setup data supports it. The example model lists streaming support.

import os
from litellm import completion

response = completion(
    model="openai/claude-sonnet-5-5",
    api_base="https://api.outrun.to/v1",
    api_key=os.environ["OUTRUN_API_KEY"],
    messages=[{"role": "user", "content": "Reply with OK only."}],
    stream=True,
)
for chunk in response:
    if chunk.choices:
        print(chunk.choices[0].delta.content or "", end="", flush=True)

Configure a local LiteLLM Proxy

Save this as config.yaml. model_name is the alias clients send to the local proxy; litellm_params.model is the actual Outrun model prefixed with openai/. The key stays in the proxy process environment.

model_list:
  - model_name: outrun-claude
    litellm_params:
      model: openai/claude-sonnet-5-5
      api_base: https://api.outrun.to/v1
      api_key: os.environ/OUTRUN_API_KEY
  - model_name: outrun-gpt
    litellm_params:
      model: openai/gpt-6.1-sol
      api_base: https://api.outrun.to/v1
      api_key: os.environ/OUTRUN_API_KEY
  - model_name: outrun-deepseek
    litellm_params:
      model: openai/deepseek-v4-pro
      api_base: https://api.outrun.to/v1
      api_key: os.environ/OUTRUN_API_KEY

Start the proxy and call it

With the proxy extra installed and OUTRUN_API_KEY available to the process, run the command below. Keep this example bound to 127.0.0.1. The cURL model value is the configured alias, not the original Outrun ID.

litellm --config config.yaml --host 127.0.0.1 --port 4000

curl http://127.0.0.1:4000/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"model":"outrun-claude","messages":[{"role":"user","content":"Reply with OK only."}]}'

Choose another model

Change only the Outrun ID after openai/ in the SDK or YAML. Use an available text/chat model with the endpoint and capabilities shown on its model page. Keep a distinct model_name for each proxy alias.

If something goes wrong