DEV Community
Follow
AI-Team-Team - A generic dynamic multi-agent collaboration framework
AI-Team-Team (ATT) allows AI agents to freely form complex hierarchical or dynamic teams. This project enables recursive spawning of child agent teams with depth limits. Teams can autonomously configure their members, define roles, and migrate within the hierarchy. A real-time ASCII map visualizes the active team topology, and a global expert directory facilitates peer discovery. ATT employs bounded ReAct loops for reasoning and a robust parser for execution. Dialogue memory is compressed to preserve crucial conversational history. LLM generation logic is delegated to a vendor-independent callback handler. A democratic voting system governs member changes, allowing for anonymous participation. A negotiation broker manages inter-team communication, and collaborative document libraries offer controlled storage. Context protection gates restrict large file reads, falling back to outline warnings and chunk requests. Tool audits intercept and vet specific tool calls before execution. SQLite state snapshots enable crash recovery by serializing system states. A supervisory team audits dialogues and escalates anomalies. Decoupled dashboards provide clear runtime event hooks for UI updates. The project is a complete Python package with implemented features and upcoming application plans.
