Preview — full styling will appear after the next deploy completes.

agentic-ai-patterns

Multi-Agent Supervisor

Supervisor routes to specialists

A supervisor LLM reads the incoming question and routes it to the most appropriate specialist agent. Each specialist has its own focused toolset and system prompt.

flowchart TD
    S([__start__]) --> V[supervisor]
    V -->|search| A[search_agent]
    V -->|deals| B[deals_agent]
    V -->|stats| C[stats_agent]
    A --> E([__end__])
    B --> E
    C --> E

The Supervisor pattern is the simplest form of multi-agent collaboration. A central supervisor LLM reads each request and decides which specialist agent should handle it. Each specialist is a complete ReAct agent with its own tools, system prompt, and domain expertise — one might search book metadata, another find deals, another compute stats.

Routing is implemented with conditional edges: the supervisor's output determines which specialist node the graph transitions to. Once a specialist finishes, execution terminates rather than returning to the supervisor (no multi-turn delegation in this variant).

This pattern excels when different types of questions require genuinely different capabilities or data sources. It keeps each agent's context focused and makes the system easy to extend — adding a new specialist only requires a new node and a routing case in the supervisor.