Agent Network: Open-Source AI Agent Collaboration Framework

What changed

The Agent Network project has been released as an open-source framework under the Apache 2.0 license. It facilitates the creation of a network of AI agents that can collaborate and delegate tasks to each other. The system supports agents powered by Claude Code, Claude Agent SDK, Codex, and Grok Build, with plans to integrate more LLMs. It utilizes MCP for agent discovery and SSE for real-time task distribution. A web-based dashboard is provided for monitoring and task assignment, and the entire system is designed to run locally, giving users control over their data.

Key Features:

• Multi-Agent Collaboration: Connects agents from various LLM providers.
• Task Delegation: Agents can discover and assign tasks to teammates.
• Local-First Architecture: Hub, Dashboard, and data storage run on user-controlled machines.
• Web Dashboard: Provides a graphical interface for managing agents and tasks.
• LLM Support: Integrates with Anthropic, OpenAI, xAI, MiniMax, DeepSeek, GLM, Kimi, Sheng, and Xiaomi MiMo models.

Installation requires Node.js version 22.13 or higher, with commands provided for setting up the hub, dashboard, and agent nodes. Default credentials are provided, with a strong recommendation to change them immediately for any public deployment.

Why it matters for builders

This release empowers builders to move beyond single-agent deployments and construct sophisticated, multi-agent systems. The ability to orchestrate different AI models for specific sub-tasks within a larger workflow offers a significant advantage in developing more capable and efficient AI applications. The open-source nature and local-first design also cater to developers prioritizing data privacy and customization.

Practical impact

Developers can now experiment with building AI teams by installing Agent Network via npm. The provided quick-start guide outlines the steps to set up a local hub and agent nodes, enabling immediate testing of inter-agent communication and task delegation. Exploring the documentation on runtime selection and multi-model integration will be crucial for tailoring the network to specific use cases.

Caveats and source limits

The source provides details on the framework's architecture, installation, and supported LLMs. However, specific performance benchmarks or independent evaluations of agent collaboration efficiency are not included. The exact capabilities of the "preview channel" features, such as Codex TUI coexistence and OpenCode integration, require further exploration within the project's documentation.