What can an AI agent do with Mistral?

A Mistral AI agent can send chat completion requests to any available model, compare outputs across model versions, create and monitor fine-tuning jobs, generate embeddings, and wire results into downstream tools like Slack, Notion, or your database, all on a schedule or triggered by an event. Write actions like starting fine-tuning jobs require your approval.

Which Mistral models can my agent use?

Your agent can use any model available on your Mistral account, including Mistral Large, Mistral Small, Codestral, and any fine-tuned models you have deployed. It can also list available models at runtime and pick the right one based on your task instructions.

Can I use this to automate fine-tuning pipelines?

Yes. You can configure a task that watches for new training data, kicks off a fine-tuning job via the Mistral API, polls for completion, and notifies you when the new model is ready, all with the plan shown to you first.

How is a Mistral AI agent different from calling the API directly?

Calling the Mistral API directly requires writing and maintaining code. A Mistral AI agent wraps that API in a reasoning layer - it decides which model to use, handles retries, formats outputs, and routes results to other tools in your stack, all based on plain-language task instructions.

Developer tools

Mistral, run by an agent.

An AI agent that runs, fine-tunes, and deploys your Mistral AI models.

Connect Mistral Book a demo

What needs your approval

Mistral never changes anything without you.

Anything that creates, edits or deletes in Mistral waits for your approval. Approve a recurring task once and it runs unattended after that.

Connecting Mistral

About 5 minutes

Auth: API key
Setup: About 5 minutes

Where to get access

Create an API key in the Mistral console at console.mistral.ai.

console.mistral.ai

Example tasks

Real work people hand to an agent that uses Mistral.

run nightly prompt regression tests across mistral model versions and post a diff summary to slack
compare mistral and openai completions on the same benchmark and log results to a spreadsheet
kick off a fine-tuning job when a new training dataset lands in cloud storage, then notify your team when it completes
generate embeddings for new support tickets and route them to the right team based on semantic similarity