Ask Mistral AI, make embeddings or read a document

The Mistral AI steps run on your company's own Mistral key. Ask a model answers a prompt with text or named fields, as the other AI steps do. Create embeddings turns text into vectors for search and matching. Read a document (OCR) reads a PDF or image — a scanned invoice, a signed delivery note — into text, page by page.

Who can do this

Workspace Admins and Editors, on every plan.

Before you start

A Mistral AI connection in Connections — see Connect Mistral AI.

Steps

Ask a model:

1. Select + where the question should be asked. In the step picker, open AI, then Mistral AI, and select Ask a model.
2. In Connection, choose the connection with your key.
3. Optional: in Model, choose a model. The list is read from Mistral with your key; left empty, the step uses Mistral's current general model, set in Bizomate. To use a model the list does not show, select Use a value and type its name.
4. Optional: in Instructions, say who the model is and how to answer — You sort customer messages for Northwind Logistics. Answer with one word only.
5. In Prompt, ask the question, with the item's values in {{ }} — Is this a quote request, a complaint or other? {{ $json.text }}
6. Under Give me back, choose Fields (the default) — named answers later steps can use one by one — or Text.
7. For Fields, add a row per answer: the name, and what goes in it — category — quote | complaint | other.
8. Select Run this step, then check Output. Pin it while you build the steps after — see Pin a step's output.

Create embeddings:

1. In the step picker, under Mistral AI, select Create embeddings, and choose the Connection.
2. In Model, keep Mistral's embedding model, or choose another from the list.
3. In Text, give the text — {{ $json.description }} — or a list of texts from an earlier step. Each text gets its own vector, in the order given.
4. Select Run this step.

Read a document (OCR):

1. In the step picker, under Mistral AI, select Read a document (OCR), and choose the Connection.
2. Under File, select Add a file and drag in the PDF or image from an earlier step — a form's upload, or an email attachment. See Pass files between steps.
3. Optional: in Pages, the pages to read — 1-3 — when you need only some of a long document.
4. Select Run this step. Each page comes back as text, with headings and tables kept as Markdown.

Output

usage holds model, tokensIn, tokensOut and seconds — for OCR, pages in place of tokens. Output shows them as the usage strip. Every test run of this step spends them again — which is what Pin is for.

Good to know

• Each item is a separate call, and a separate cost. Twenty items, twenty calls.
• Mistral bills your account, not Bizomate. OCR is charged by the page.
• Files pass between steps up to 250 MB; Mistral may refuse a very large document — read fewer Pages at a time.
• Fields are always all filled. The model must give every field named; a field with | choices gets one of them.
• For a busy account, set If this step fails to try again — see Make a step try again when it fails.

If something goes wrong

Related

• Connect Mistral AI
• Ask an AI model with Claude, OpenAI or Gemini
• Pass files between steps
• Pin a step's output