mistralai/client-python

What is the Mistral Python Client?

The repository provides an official Python SDK for Mistral AI’s cloud services – chat‑completion, embeddings, audio, file handling, agents, and a host of other resources. It lets developers call Mistral’s REST APIs from Python code (both synchronously and with asyncio) without having to hand‑craft HTTP requests.

Key Features (as described in the README)

Area	What you can do with the SDK
Chat completions	Generate text responses from models such as `mistral-large-latest`. Supports streaming and custom response formats.
Embeddings	Obtain vector embeddings for one or many input strings via `mistral-embed`.
File uploads	Upload files to a workspace (e.g., for fine‑tuning or retrieval‑augmented generation).
Agents	Call agent endpoints that maintain stateful interactions, with both sync and async APIs.
Audio	Speech synthesis (`Audio.Speech`) and transcription (`Audio.Transcriptions`), including streaming transcription via Server‑Sent Events.
Batch jobs	List, create, cancel, and delete long‑running batch jobs.
Beta resources	Manage agents, connectors, conversations, libraries, and documents (create, list, update, delete, pagination, etc.).
Cloud provider wrappers	Ready‑made clients for Azure AI and Google Cloud (Vertex AI) that automatically handle endpoint construction, authentication, and token refresh.
Utilities	Built‑in pagination, retries, error handling, custom HTTP client injection, telemetry hooks, and IDE‑friendly type hints.

Getting Started

Obtain an API key from the Mistral console (or from Azure/GCP if you’re using those wrappers). Export it as MISTRAL_API_KEY (or the provider‑specific env vars).
Install the package – any of the following works:
```
# uv (fast installer)
uv add mistralai

# pip (default)
pip install mistralai

# poetry
poetry add mistralai
```
For agent‑related features add the extra: pip install "mistralai[agents]" (requires Python 3.10+).

Write a tiny script (sync example):

from mistralai.client import Mistral
import os

with Mistral(api_key=os.getenv("MISTRAL_API_KEY")) as client:
    resp = client.chat.complete(
        model="mistral-large-latest",
        messages=[{"role": "user", "content": "Who is the best French painter?"}],
        stream=False,
        response_format={"type": "text"},
    )
    print(resp)

The same pattern works with async with Mistral(...): and the *_async methods.

When Would You Use This?

Building LLM‑powered applications (chatbots, assistants, RAG pipelines) that need a reliable, typed Python interface.
Integrating Mistral models into existing Python services without dealing with raw HTTP.
Leveraging Mistral’s extended ecosystem – audio generation, transcription, batch processing, or the newer agents and connectors features.
Deploying on Azure or Google Cloud – the SDK includes thin wrappers (MistralAzure, MistralGCP) that automatically handle the provider‑specific authentication and endpoint construction.

How It Is Structured

mistralai.client.Mistral – the core client exposing sub‑objects (chat, embeddings, files, agents, audio, etc.).
Provider‑specific modules (mistralai.azure.client, mistralai.gcp.client) that subclass the core client with preset base URLs and auth flows.
Extensive documentation in the docs/ folder for each resource, plus a examples/ directory that can be run with uv run.

TL;DR

The Mistral Python Client is a fully‑featured, officially supported SDK that wraps Mistral AI’s cloud APIs. Install it via pip (or uv/poetry), set your API key, and you can start making chat, embedding, audio, file, and agent calls directly from Python—both synchronously and asynchronously. It also includes ready‑made adapters for Azure AI and Google Cloud, making it a convenient entry point for any Python developer working with modern LLM services.