Mistral announced its new Agents application programming interface (API) on Tuesday. The new API comes equipped with connectors for code execution, web search, image generation, as well as support for Model Context Protocol (MCP) tools. These capabilities are powered by several artificial intelligence (AI) models, including the recently released Devstral as well as Black Forest Lab’s Flux 1.1 [pro] Ultra. Mistral is also adding enhanced memory functions to let models complete complicated tasks and support for AI agent creation via its API.
Mistral’s Agents API Will Let Developers Create Multi-Agent Workflows
In a newsroom post, the Paris-based AI firm detailed its new developer-focused product. The company stated that the Agents API addresses the limitations of large language models (LLMs) in executing actions and maintaining context across subtasks or multiple tasks. The new offering is being made available alongside Mistral’s existing Completion API.
Mistral highlighted that the Agents API features a dedicated framework that simplifies both agent creation and using multi-agent workflows. The API makes it easy to create an agent and equip it with built-in connectors. These connectors are essentially tools that can be deployed to let AI agents take various actions.
The code execution tool allows developers to create agents that can execute Python code in a virtual sandboxed environment. Similarly, the image generation connector lets agents generate images for various use cases. The tool is powered by the Flux 1.1 [pro] Ultra model. Then there is the document library tool, which enables agents to access documents from Mistral Cloud.
Apart from these, two more powerful connectors have been added. First is web search, which allows an agent to browse the Internet to find real-time information, credible news, and other data. Mistral says agents with web search capabilities display higher efficiency in performance.
Finally, the Agents API software development kit (SDK) also supports tools that are built on Anthropic’s MCP. Notably, the protocol provides a standardised way for agents to connect with external data hubs.
Beyond these built-in connectors, the Mistral API also comes with two new features — improved memory and creation of orchestrator agents. With improved memory, the API retains the context of the conversations, so that the agents can retrieve information from past conversations when taking on new tasks. Developers will also not have to track conversation history as they can both view past conversations and continue any of them.
Finally, developers will also be able to create orchestrator agents with the API. These can be understood as directors who delegate subtasks to relevant agents and supervise the completion of the task. Mistral explained that once an agent is created, developers can define which agents they can hand off tasks to, to seamlessly design the orchestrator.
