Swarm Agent Framework

What it does

Swarm is a robust framework written in Rust for building and managing complex networks of specialized AI agents. It provides a central nervous system for digital workforces, employing a 'Conductor' model (Planner and Executor agents) to orchestrate tasks through dynamic or static workflows. It features a built-in LLM-as-a-Judge evaluation system to enable self-correcting behavior and an Agent Factory for programmatic agent instantiation.

Tools

• Agent Factory: Programmatically launches specialist agents with specific MCP runtimes.
• Planner Agent: Generates step-by-step execution plans based on high-level goals.
• Executor Agent: Coordinates with domain agents and tools to carry out plans.
• Evaluation Service: Uses LLM-as-a-Judge to assess success and trigger plan refinement.
• MCP Runtime: Bridges agents to external tools and data sources.

Installation

Since this is a framework, build from source:

git clone https://github.com/fcn06/swarm.git
cd swarm
cargo build --release

Configure LLM keys (Groq/Gemini) via environment variables: LLM_A2A_API_KEY, LLM_MCP_API_KEY, LLM_JUDGE_API_KEY, LLM_PLANNER_API_KEY.

Supported hosts

• Claude Desktop (via MCP runtime integration)