Mistral AI launches Agents API

Mistral AI, a French AI startup, is launching its Agents application programming interface (API) for enterprise customers and software developers. The API lets developers add autonomous generative AI capabilities to their applications, using the Mistral Medium 3 model.

The platform is designed to be plug and play, allowing customization to get AI agents running for enterprise and developer workflows. According to Mistral’s Head of Developer Relations, Sophia Yang, the Agents API lets enterprises use AI in more practical ways.

This release is designed to work with Mistral’s Chat Completion API, and focuses on agentic orchestration, built-in connectors, persistent memory, and the ability to coordinate multiple AI agents for complex tasks.

The Agents API gives developers tools to create AI agents that can perform real-world tasks, manage interactions, and orchestrate multiple agents. It has built-in connectors and supports MCP tools, connecting agents to external resources like APIs, databases, user data, and documents.

One feature is the integration of web search as a connector, which improves performance on tasks needing up-to-date information. In testing on the SimpleQA dataset, Mistral Large’s accuracy increased from 23% to 75% when web search was enabled. Mistral Medium showed an improvement, increasing from 22.08% to 82.32%.

The Agents API’s conversation system maintains context throughout interactions. Developers can start conversations without losing the thread, and conversation history is stored for future use. The API supports streaming output, enabling real-time updates.

A key capability is its ability to coordinate multiple agents. Developers can create workflows, assigning tasks to agents and enabling handoffs. This allows enterprises to deploy AI agents that work together to solve problems.

Key Benefits for Engineers

The Mistral Agents API can help senior-level engineers dynamically orchestrate agents and integrate data sources, which can help with quick iteration and performance tuning. Built-in connectors can reduce the need for integrations. Stateful conversation management and real-time updates also align with AI orchestration and deployment needs.

MCP tool support allows agents to access data from APIs and systems, enhancing operational efficiency.

Le Chat Enterprise

The Agents API launch follows the launch of Le Chat Enterprise, an AI assistant platform for productivity and data privacy. Le Chat Enterprise is powered by the Mistral Medium 3 model, which performs well at a lower computational cost. Mistral Medium 3 is strong in software development tasks and shows performance in multilingual and multimodal scenarios.

Le Chat Enterprise has features such as data sovereignty, hybrid deployment, and access controls. The platform consolidates AI functionality, enabling customization, integration, and control over deployment and data security.

Open Source

Mistral Medium 3 is a proprietary model, requiring access through Mistral’s platform, APIs or partners and is no longer available under an open-source license. The Agents API itself also follows a proprietary framework, managed by Mistral, with access via subscription and API calls. Pricing for the Agents API aligns with Mistral’s suite of models and tools.

Mistral AI positions its Agents API as the foundation for enterprise-grade agentic platforms. The focus on enterprise features, workflows, and integrations makes this API an option for businesses seeking AI capabilities. Mistral encourages developers to explore the documentation and demos.