How to use chat.ai.e-infra.cz via API

Generating an API token

Log in to https://chat.ai.e-infra.cz/ via browser using your Metacentrum credentials. If you don't have one, here is how to get access.
At chat.ai.e-infra.cz, on the top right corner, click on the profile icon and open Settings.
Navigate to Account and click on API keys (display).
Ignore JWT token and select API key and either generate new or display existing.
Copy the generated API key and store it securely.
Use this key in API requests to authenticate and access Open-WebUI services.

API endpoint is: https://llm.ai.e-infra.cz/v1/.

Listing available models

As the models are named differently in the web UI, it is necessary to list the ones that are available via the API. This can be done using the following command:

export EINFRA_AI_TOKEN="..."
curl -H "Authorization: Bearer $EINFRA_AI_TOKEN" https://llm.ai.e-infra.cz/v1/models | jq .data[].id

Where EINFRA_AI_TOKEN refers to the e-infra API key acquired in previous steps. It is also a good way of checking whether the token is valid. The output should be similar to this:

"gpt-oss-120b"
"deepseek-r1"
"qwen3-coder"
"qwen2.5-coder:32b-instruct-q8_0"
"medgemma:27b-it"
...

Inference

As the E-Infra uses OpenAI-compatible server, you can use any Python framework that is capable of HTTP requests. Several options for inspiration are:

openai-python: the official implementation for OpenAI API.
PydanticAI: an extensive toolkit that supports advanced applications.
requests: low-level approach, recommended when you want custom behavior.

Example using `openai-python`

import os
import openai

MODEL = "deepseek-v3.2-thinking"
BASE_URL = "https://llm.ai.e-infra.cz/v1/"

client = openai.OpenAI(
    # using environment variables are one of the safer ways to pass your API keys
    api_key=os.environ["EINFRA_AI_TOKEN"],
    #openai uses chatgpt url as default - we need to override it here
    base_url=BASE_URL,
)

response = client.chat.completions.create(
    #here you paste the selected model
    model=MODEL,
    messages=[
        {"role": "system", "content": "Talk like a pirate."},
        {
            "role": "user",
            "content": "What is a mast?",
        },
    ],
)

print(response.choices[0].message.content)

References

Official documentation: https://docs.cerit.io/en/docs/ai-as-a-service/chat-ai