Wanman: Open-source agent matrix network with JSON-RPC communications (github.com)

The newly announced wanman is an open-source local-mode agent matrix framework designed to manage multiple AI agents, specifically Claude Code and Codex, on a user's machine. Coordinated through a JSON-RPC supervisor, wanman allows users to observe these agents—each functioning as a distinct subprocess—while minimizing direct interaction. The framework's name, inspired by a Japanese term for a one-man train, reflects its design philosophy of allowing the user to assume a passive role as the agents autonomously execute tasks such as development, marketing, and feedback management. This development is significant for the AI/ML community as it provides a robust, scriptable, and reproducible environment for managing AI workflows. By isolating each agent into its own workspace and utilizing an asynchronous messaging system, wanman enhances collaboration and task orchestration among agents. Furthermore, it includes features for dynamic role extraction, self-evolving skills, and high-concurrency task execution, promoting scalability. The CLI-first approach ensures that everything is easily accessible and manageable through code, simplifying the integration and use of AI agents in various applications.