French startup Mistral AI introduced Agents API — a framework for creating autonomous AI agents with built-in connectors, persistent memory, and orchestration capabilities. Developers can create unlimited agents and build pipelines for interaction between them.
Agents API includes five built-in connectors.
- Code Execution runs Python code in a secure environment for mathematical calculations, data visualization, and scientific computing;
- Image Generation uses Black Forest Lab FLUX1.1 [pro] Ultra for image creation;
- Document Library works with documents from Mistral Cloud with integrated RAG functionality;
- Web Search collects information from internet sources;
- MCP Tools – an open standardized protocol for integration with external systems, APIs, and databases.
Quantitative Results
Web search integration demonstrates significant performance improvements. In SimpleQA tests, Mistral Large accuracy increased from 23% to 75%, Mistral Medium — from 22.08% to 82.32% when web search capabilities were enabled.
Context Management
The API uses a stateful conversation system with context preservation between interactions. Developers can create chats with specific agent identifiers or direct model access. The framework supports conversation branching and streaming output for real-time updates.
Agent Orchestration
The key feature is coordination of multiple agents through dynamic orchestration. Task handoffs between agents enable solving complex problems through coordination of several agents. Developers can create specialized agents for specific functions and configure their interaction. For example, a financial agent passes data search tasks to a web agent, then receives results back for analysis. A single user request can initiate a chain of actions across multiple agents, where each processes its part of the task.
Practical Applications
Mistral published agent examples for various sectors:
- Developer agent with GitHub integration for automated development;
- Agent for managing project tasks in Linear;
- Analyst agent for extracting metrics and compiling reports;
- Travel agent for trip planning;
- Nutrition agent for diet management.
Pricing
API call prices: Mistral Medium 3 — $0.4/$2 per million input/output tokens, Web Search — $30 per 1,000 calls, Code Execution — $30 per 1,000 calls, Image Generation — $100 per 1,000 images. Document Library is included in Team and Enterprise plans.
Strategic Positioning
Agents API follows the launch of Le Chat Enterprise and serves as the backbone for enterprise-grade agent platforms. The shift to proprietary solutions reflects Mistral’s focus on corporate clients, moving away from previous open-source approaches.
For enterprise teams, the API creates a streamlined approach to deploying AI solutions with simplified integration. Built-in connectors reduce the need for developing custom tools, allowing focus on model fine-tuning rather than infrastructure development.
The API is positioned as a comprehensive solution for creating AI agents capable of performing real-world tasks beyond traditional text generation, with documented performance improvements and integration flexibility.
