Multi-agent
tnsai-coordination runs several agents on one task: group topologies, workflows, messaging, delegation, supervision and decision patterns such as councils, debates and judges. Auto team is in tnsai-intelligence. If you are not sure which shape fits, read Group topologies and then Topology selection.
| Page | Covers |
|---|---|
| Group topologies | The eight group shapes and their builders |
| Topology selection | Which topology, protocol or solver to pick, and when team, hierarchy or pipeline is a trap |
| Workflows | DAGs, pipelines, sagas, branches, loops and execution plans |
| Fan-out and agent pools | Sending one task to several agents and pooling workers |
| Handoffs, streams and choreography | Passing work and context from one agent to the next |
| Shared state and trust | The shared blackboard, trust scores and trust-based routing |
| Supervision and safeguards | Supervisors, restarts and limits on a running group |
| Recursive task solver | Splitting a task into a tree of verified sub-tasks |
| Council and voting | Councils, voting strategies and decisions |
| Debate and peer review | Agents arguing positions and reviewing each other |
| Judge | A quality gate that approves, retries or rejects one agent's output |
| Negotiation | Contract-net, auctions and bargaining |
| Auto team | An LLM that composes a group for a task |
| Messaging | Agent-to-agent messages, routing and delivery |
| Protocols | Performatives, conversation protocols and transports |