Enjambre – a durable kernel for swarms of AI agents (Python, MCP) (github.com)

Enjambre, a new operating system designed for managing swarms of AI agents, has been announced. It enables seamless collaborative work among diverse AI models and scripts, such as Claude Code and Codex, without the risk of misinformation. This framework includes essential components such as a kernel, a queue, permission gates, and shared memory. Enjambre provides a user-friendly demo that illustrates real-world applications, like a scripted editorial process, showcasing key features like task retries, permissions enforcement, and a visible decision-making router. The significance of Enjambre lies in its innovative approach to swarm management, addressing common failures in these systems—such as blocked resources and task dependencies—that can lead to inefficiencies and errors. Enjambre employs mechanisms like durable queues, atomic task management, and a unique memory graph that enhances task verification and recovery. This ensures a higher level of reliability and transparency, making it easier to oversee agent interactions and improve the success rates of coordinated tasks. By allowing operation on various platforms (Linux, macOS, Windows) and requiring minimal setup with Python, Enjambre aims to empower developers to leverage distributed AI agents more effectively, fostering advancements in collaborative AI workflows.