Use Mistral Models for Text, Embeddings and Images

The Mistral actions let a workflow call Mistral language and vision models with your own API key. They are in beta, so test before relying on them for customer-facing messages.

Connect your account

Add any Mistral action to a workflow and click Connect Now, then approve access. You can also go to Settings > Integrations > Mistral AI and click Connect. Mistral has no triggers, so start the workflow with a standard one such as Form Submitted or Customer Replied.

The three actions

• Create Chat Completion writes replies, summaries, rewrites or classifications. Fields: Action Name, System Prompt, Prompt, Model (Large, Medium, Small or Ministral) and tuning options for length and style.
• Create Embeddings converts text into vectors with the mistral-embed model, handy for semantic matching, classification and clustering.
• Analyze Image (Vision) reads one or more images, such as receipts, screenshots, business cards or form uploads, using models like Pixtral.

Set up a chat completion

1. Add the action from the Apps tab in the builder.
2. Name it, then write a System Prompt that defines role and tone.
3. Write the Prompt and insert merge fields from the contact.
4. Choose a model and save.
5. Insert the output variable into a later message or field update.

Good to know: Pick a smaller model for simple sorting tasks and a larger one for writing, since usage is billed by Mistral.