🔬

Early preview: This API is in early preview. Fill out this form to request early access.

The Models API is OpenAI-compatible: request bodies, parameters, and response objects match the OpenAI HTTP API, so you can reuse existing OpenAI tutorials and SDKs by swapping the base URL and API key.

Before you begin

• A monday.com Pro or Enterprise workspace with AI features available for your use case.
• Early access to the Models API preview (request via the form linked above).

1. Get your API token

Use a monday.com personal API token as the credential for the Models API (not an OpenAI API key). How to create and copy your token, regenerate it, and what it can access is documented in Authentication.

If you authenticate with an app token instead, configure scopes and tokens as described in the apps documentation so the token can use AI features. The app associated with the token must include the ai:consume scope.

2. Point your client at the Models API

Set:

If you use the OpenAI SDKs, set baseURL / base_url to the value above and apiKey / api_key to your monday token.

3. Send a test request

cURL (chat completion)

curl -sS -X POST "https://api.monday.com/platform-ai-gateway/openai/v1/chat/completions" \
  -H "Authorization: Bearer <MONDAY_API_TOKEN>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "monday-standard",
    "messages": [{ "role": "user", "content": "Reply with the word: ok" }]
  }'

We recommend monday-standard (or monday-fast / monday-powerful) for tests. See Recommended chat model aliases. You can also pass a direct model ID from the supported models list.

Supported capabilities

Together, these cover text, embeddings, images, and audio through one API.

Chat model aliases (recommended)

For /chat/completions, pass one of the monday-* aliases instead of hard coding a vendor model ID whenever you can:

• monday-fast → claude-haiku-4-5
• monday-standard → claude-sonnet-4-6 (also used when model is omitted)
• monday-powerful → claude-opus-4-6

Why aliases: They are more stable for your product. monday controls how each alias maps to an underlying model, so improvements (for example switching to a newer Sonnet) do not require you to redeploy new model strings everywhere. Aliases also match our pricing policy: there is no extra surcharge on top of underlying usage when you use them, unlike direct vendor IDs. See Supported models and Pricing and metering.

Using official OpenAI SDKs

Point the client at the Models API base URL and use your monday token as the API key. The request and response types match what the OpenAI SDKs expect, so you can follow any tutorial written for OpenAI by swapping base URL and API key.

• Full method list and parameters: OpenAI API reference
• Cookbook and patterns: OpenAI documentation — use the same request shapes; validate model IDs against Supported models because not every OpenAI model name is available on the Models API.

👍

Tip

Some OpenAI SDKs require an explicit model argument even when the upstream server has a default. When you must pass a value, use monday-standard (or another monday-* alias) rather than a raw vendor ID. If you rely on server-side defaults via REST, the Models API defaults to monday-standard.

Behavior notes (preview)

Expect the Models API to evolve during early preview. A few practical differences from calling OpenAI directly are worth planning for:

• Images: Responses may return image data as b64_json rather than a hosted url—decode base64 in your app.
• Errors: Some HTTP errors (for example authentication failures) may use a platform-specific JSON shape instead of the OpenAI error object. Handle both raw HTTP failures and structured error bodies.
• Models: Unsupported model names can surface as generic upstream errors—prefer the supported models list and pin explicit IDs in production.

Metering

During early preview, there is no separate fee to use the Models API itself. Usage draws down monday AI tokens based on underlying provider cost, plus alias vs direct ID rules and preview rounding described in Pricing and metering.

Check response headers (such as x-monday-credits-used) while integrating so your app can log or display usage consistently.

See also