Skip to main content

Mistral AI and xAI Grok

Mistral AI and xAI (Grok) are direct providers. You paste one API key and the gateway calls the vendor's own API. Your code keeps using a Quilr key and the standard gateway endpoints, so there is no Mistral- or xAI-specific path or model prefix. Clients on other formats (Responses, Anthropic Messages, Gemini, Bedrock) reach both through any-to-any formats.

Provider typeConsole tile and APIUpstreamWhat one key serves
mistralMistral AI › Chat completionshttps://api.mistral.ai/v1Chat, embeddings, transcription, moderation, document OCR, Codestral fill-in-the-middle (FIM). Files and batch jobs when enabled.
xaixAI › Chat completionshttps://api.x.ai/v1Chat, Grok Imagine image generation and image edits. Files, batches and video generation when enabled.
xai_responsesxAI › Responseshttps://api.x.ai/v1The Responses API, which xAI recommends for new integrations.

Add the provider​

  1. Open Settings > Models > Models and click Add your own models. The same picker appears when you create an app or edit an app's providers.
  2. Type mistral, grok or xai in Search provider types (placeholder: Search by vendor or API). Search is not case-sensitive and matches the vendor name, tagline, description and API names, so queries such as ocr, image or batch also find these tiles.
  3. Select the Mistral AI or xAI tile. For xAI, choose Chat completions or Responses under Which API does the gateway talk to?. Mistral has one API, so there is no choice.
  4. Keep or edit the suggested Provider label and paste the API key. There is no Base URL, region or API version field: the upstream URL is fixed by QuilrAI, and any base_url sent through the Management API is ignored.
  5. Click Check key and list models, select the models, set pricing and save. See Providers and Models for the full flow.
FieldAPI nameRequired
Provider labelprovider_labelYes. Letters, numbers, spaces, _, - and ..
API keyapi_keyYes. In edit mode, leave it blank to keep the saved key.
ModelsmodelsAt least one.

Model discovery​

  • Mistral lists models from /v1/models and sorts each one into chat, embedding, moderation, OCR, transcription or FIM from Mistral's capability flags. Archived models are hidden.
  • xAI lists language models, plus image and video generation models where your key can reach them. xai_responses lists language models only.
  • Selected chat models get a short test call when you validate them.

Enable every model you call, on every API. A Mistral embedding or OCR model has to be selected on the provider just like a chat model.

Call the gateway​

Use a regional base URL and your Quilr key (Authorization: Bearer <Quilr key>).

What you wantEndpointMistralxAIExample model
Chat/openai_compatible/v1/chat/completions✓✓mistral-large-latest, ministral-8b-latest, grok-4.20-0309-non-reasoning
Responses/openai_responses/v1/responsesTranslated to chat✓ (xai_responses, or xai by endpoint switch)grok-4.20-0309-non-reasoning
Anthropic Messages, Gemini, Bedrock ConverseTheir native pathsTranslatedTranslated-
Embeddings/openai_compatible/v1/embeddings✓-mistral-embed
Speech to text/openai_compatible/v1/audio/transcriptions✓-voxtral-mini-latest
Moderation/openai_compatible/v1/moderations✓-mistral-moderation-latest
Document OCR/openai_compatible/v1/ocr✓-mistral-ocr-latest
Fill-in-the-middle/openai_compatible/v1/completions✓-codestral-latest
Image generation/openai_compatible/v1/images/generations-✓grok-imagine-image
Image edits/openai_compatible/v1/images/edits-✓ (JSON body)grok-imagine-image
Files, batches/openai_compatible/v1/files, /batches✓✓-
Video generation/openai_compatible/v1/videos-✓-

Moderation, OCR, FIM, images, files, batches and videos are endpoint families. Files, batches and videos are off by default.

Examples​

# Mistral chat
curl https://guardrails.quilr.ai/openai_compatible/v1/chat/completions \
  -H "Authorization: Bearer $QUILR_KEY" -H "Content-Type: application/json" \
  -d '{"model": "ministral-8b-latest", "messages": [{"role": "user", "content": "Hello"}], "max_tokens": 128}'

# Codestral fill-in-the-middle (streaming works too)
curl https://guardrails.quilr.ai/openai_compatible/v1/completions \
  -H "Authorization: Bearer $QUILR_KEY" -H "Content-Type: application/json" \
  -d '{"model": "codestral-latest", "prompt": "def add(a, b):\n    return", "suffix": "\n\nprint(add(1, 2))", "max_tokens": 32}'

# Mistral document OCR (Mistral request shape)
curl https://guardrails.quilr.ai/openai_compatible/v1/ocr \
  -H "Authorization: Bearer $QUILR_KEY" -H "Content-Type: application/json" \
  -d '{"model": "mistral-ocr-latest", "document": {"type": "document_url", "document_url": "https://example.com/invoice.pdf"}}'

# Mistral transcription
curl https://guardrails.quilr.ai/openai_compatible/v1/audio/transcriptions \
  -H "Authorization: Bearer $QUILR_KEY" \
  -F file=@hello.wav -F model=voxtral-mini-latest

# xAI image generation
curl https://guardrails.quilr.ai/openai_compatible/v1/images/generations \
  -H "Authorization: Bearer $QUILR_KEY" -H "Content-Type: application/json" \
  -d '{"model": "grok-imagine-image", "prompt": "A plain white square.", "n": 1}'
from openai import OpenAI

client = OpenAI(base_url="https://guardrails.quilr.ai/openai_responses/v1", api_key=QUILR_KEY)
resp = client.responses.create(model="grok-4.20-0309-non-reasoning", input="Say hello", max_output_tokens=128)
print(resp.output_text)

If an app has several providers that serve the same model, pick one with provider_label or X-Provider-Label. See Selecting a Provider.

Provider behavior​

TopicMistral AIxAI
Output limitSend max_tokens. Mistral has no max_completion_tokens and no stream_options.max_completion_tokens is accepted. The gateway requests streaming usage for you.
Rejected fieldsUnsupported fields come back as Mistral's own HTTP 422.Reasoning models reject presence_penalty, frequency_penalty and stop; the error is passed through.
StreamingChat and FIM. Reasoning streams that send delta.content as a list of parts are supported.Chat and Responses.
ModerationThe response gets OpenAI's per-result flagged field (true when any category is true).-
FIMseed becomes random_seed. n, best_of, echo, logprobs, logit_bias, penalties and user are dropped. The response uses OpenAI's choices[].text.Not available: xAI rejects /v1/completions for every current model.
Images-size, quality, style, background, moderation, output_compression, output_format and user are dropped. Image edits take a JSON body, not multipart.
Responses-Create, retrieve and delete. cancel and input_items return 400 operation_not_supported.
Not availableAudio translations.Legacy completions.

Dropped fields are recorded on the request log.

Guardrails, routing and limits​

  • Chat, Responses, embeddings and transcription get the same guardrails as any other provider.
  • Moderation, OCR, FIM, images, files, batches and videos are not scanned by DLP. Access controls, model allow-lists, rate and token limits and quotas still apply. See Endpoint Families.
  • Routing groups: mistral and xai are own-family members of Chat Completions groups; xai_responses of OpenAI Responses groups. See Routing and Fallbacks.

On this page