Skip to main content

Supported API endpoints

The API is accessible via HTTPS and is compatible with the OpenAI API. The base URL is https://llm.aihosting.mittwald.de. Many applications require specifying the version as well. In this case, the full base URL with versioning should be used: https://llm.aihosting.mittwald.de/v1.

Every interaction with the API requires an Authorization header with a valid API key. This key can be created in mStudio.

In our examples, we use curl, as it is the simplest and quickest way to test. For production use, we recommend using frameworks and libraries that support OpenAI. To run the following examples please setup your API key as an environemt variable first, e.g.

export APIKEY=sk-…

/v1/models​

This endpoint returns a list of available models.

curl -i -X GET https://llm.aihosting.mittwald.de/v1/models \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $APIKEY"

It returns a dictionary containing a list of available models. The id within this list can be used in the model field for subsequent API routes.

/v1/chat/completions and /v1/completions​

This route allows content to be sent to the LLM in chat format. The endpoint supports streaming for the content generated by the LLM.

curl -i -X POST https://llm.aihosting.mittwald.de/v1/chat/completions \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $APIKEY" \
    -d '{
        "model": "Mistral-Small-3.2-24B-Instruct",
        "messages": [
            {
                "role": "user",
                "content": "Moin and hello!"
            }
        ]
    }'

The model parameter requires a valid model name, which can be retrieved via the /v1/models route. Additional model parameters such as temperature, top_p, or top_k can be provided. Recommended settings for these can be found in the model descriptions if they differ from the defaults. Which extended parameters influence the response depends on the model.

To receive a streamed response, the option stream: true must be set.

curl -i -N -X POST https://llm.aihosting.mittwald.de/v1/chat/completions \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $APIKEY" \
    -d '{
        "model": "Mistral-Small-3.2-24B-Instruct",
        "messages": [
            {
                "role": "user",
                "content": "Moin and hello!"
            }
        ],
        "stream": true,
        "temperature": 0.15,
        "top_k": 10,
        "top_p": 0.5
    }'

/v1/responses (experimental)​

This endpoint provides access to the Responses API, following the structure of the OpenAI Responses API. As the route is still experimental, not all features are fully supported. Some extended parameters may behave differently compared to the OpenAI Responses API.

Structured output such as schema-based JSON cannot currently be enforced with this endpoint. If structured output is required, the /v1/chat/completions endpoint should be used instead.

curl -i -X POST https://llm.aihosting.mittwald.de/v1/responses \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $APIKEY" \
    -d '{
        "model": "Mistral-Small-3.2-24B-Instruct",
        "input": "Explain the purpose of this experimental endpoint."
    }'

The response contains the generated text and metadata. Since the endpoint is still under development, some fields may be missing or implemented differently.

To receive a streamed response, set stream: true:

curl -i -N -X POST https://llm.aihosting.mittwald.de/v1/responses \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $APIKEY" \
    -d '{
        "model": "Mistral-Small-3.2-24B-Instruct",
        "input": "Send a short streamed message.",
        "stream": true,
        "temperature": 0.2,
        "top_p": 0.5,
        "top_k": 20
    }'

Streaming is supported, but certain functions or parameters may be limited due to the experimental status of the endpoint.

/v1/embeddings​

This route allows you to generate embeddings for texts.

curl -i -X POST https://llm.aihosting.mittwald.de/v1/embeddings \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $APIKEY" \
    -d '{
        "model": "Qwen3-Embedding-8B",
        "input": "An important document"
    }'

Depending on the embedding model used, additional supported parameters such as dimensions can be submitted. However, this is not applicable to Qwen3-Embedding-8B.

/v1/audio/transcriptions​

This endpoint allows you to transcribe audio files into text using Whisper models.

curl -X POST https://llm.aihosting.mittwald.de/v1/audio/transcriptions \
    -H "Authorization: Bearer $APIKEY" \
    -F "file=@/path/to/audio.mp3" \
    -F "model=Whisper-Large-V3-Turbo"

The file parameter should contain the audio file to be transcribed. Supported formats include MP3, OGG, WAV, and FLAC. Maximum file size is 25 MB. The model parameter specifies which Whisper model to use for transcription.

Additional optional parameters can be provided:

  • language: The language of the input audio (ISO-639-1 format). If not specified, German ("de") will be assumed by default. It is strongly recommended to always set this parameter explicitly for best accuracy.
  • temperature: Sampling temperature, recommended value is 1.0
  • response_format: Format of the response (json or text). Other formats like srt, vtt, and verbose_json are not currently supported.
curl -X POST https://llm.aihosting.mittwald.de/v1/audio/transcriptions \
    -H "Authorization: Bearer $APIKEY" \
    -F "file=@/path/to/audio.mp3" \
    -F "model=Whisper-Large-V3-Turbo" \
    -F "language=de" \
    -F "temperature=1.0" \
    -F "response_format=json"

Important notes:

  • Translation functionality (to_language parameter) is not supported
  • For best results, always specify the language parameter explicitly
  • Segment large audio files into chunks smaller than 25 MB if needed