Mistral is excited to introduce the new Agents API, a major leap forward in making AI not just intelligent, but truly useful, action-oriented, and ready to solve real-world problems.

Traditional language models have always been great at generating text. But when it comes to performing tasks, remembering conversations, or coordinating actions, they fall short. That’s exactly what the new Agents API is built to change.

By combining Mistral’s powerful models with,

• Built-in tools for running code, searching the web, generating images, and more
• Persistent memory to maintain context across conversations
• Agent orchestration, allowing multiple agents to work together

This new API gives developers and enterprises the power to build smart, capable AI systems that go far beyond simple chatbots.

What You Can Build with the Agents API?

The Agents API unlocks endless possibilities across industries. Here are just a few examples.

GitHub Coding Assistant: An agent that writes and manages code through GitHub, powered by DevStral. It oversees the whole dev process with full access to repositories, automating software tasks end-to-end.

Linear Tickets Assistant: An intelligent coordinator that converts meeting transcripts into PRDs and creates actionable Linear issues, all while tracking deliverables.

Financial Analyst Agent: A financial assistant that pulls live data, performs analysis across multiple MCP servers, and securely stores insights.

Travel Planner: A travel agent that helps users plan trips, book stays, and manage itineraries.

Nutrition Coach: A personal diet companion that logs meals, sets goals, and offers meal suggestions, all while keeping track of progress.

Built-In Tools You Can Use

Agents can use a variety of built-in connectors and MCP tools to do their jobs more effectively.

Code Execution

Run Python in a safe environment for tasks like calculations, plotting, and scientific analysis.

Image Generation

Use FLUX1.1 [pro] Ultra to create educational visuals, marketing graphics, or digital art.

Document Library

Pull content from user-uploaded docs in Mistral Cloud with built-in RAG (retrieval-augmented generation).

Web Search

Access the latest info from the web, news, and more. Performance improves dramatically.

• Mistral Large: 23% → 75%
• Mistral Medium: 22.08% → 82.32%

MCP Tools

MCP (Model Context Protocol) lets agents access APIs, databases, documents, and other live systems, perfect for dynamic, real-time tasks.

Memory and Conversation Context

Conversations are now stateful, meaning agents can remember past chats and keep context.

• Start with a specific agent or model
• Pick up from any past conversation
• Branch out into new paths
• Stream live responses as they work

This gives users a seamless and consistent AI experience, more like working with a human assistant.

Agent Collaboration and Handoffs

One of the most powerful features? Agentic workflows.

You can create multiple agents, each with its own roles and tools, and define task handoffs between them.

Example: A financial agent may pass data fetching to a search agent, then hand it off again to a calculator agent. All this happens behind the scenes, making the AI experience smoother and smarter.

Ready to Start?

Dive into the docs, create your first agent, and start experimenting. With built-in tools, memory, and orchestration, the Mistral Agents API is designed to turn ideas into action, one smart agent at a time.