Multi-Agent Systems: Orchestrating Agents at Scale

    Who this is for:

    Architecture / Concept Overview: Multi-Agent Systems: Orchestrating Agents at Scale

    A multi-agent system uses a supervisor pattern to coordinate specialised agents, each responsible for a distinct domain.

    %%{init: {"theme":"base","themeVariables":{"background":"#0B0E14","primaryTextColor":"#E0E6ED","lineColor":"#5D6470","darkMode":true,"primaryColor":"#2E4A4A","secondaryColor":"#374151","secondaryTextColor":"#E0E6ED","tertiaryColor":"#111827","tertiaryTextColor":"#E0E6ED","edgeLabelBackground":"#1f2937"}}}%% flowchart LR classDef source fill:#3F4B59,stroke:#9CA3AF,stroke-width:2px,rx:8,ry:8,color:#E0E6ED classDef ingestion fill:#5A4B36,stroke:#C9A86B,stroke-width:2px,rx:8,ry:8,color:#E0E6ED classDef processing fill:#535072,stroke:#8E82B4,stroke-width:2px,rx:8,ry:8,color:#E0E6ED classDef storage fill:#2E4A4A,stroke:#5FAFA8,stroke-width:2px,rx:8,ry:8,color:#E0E6ED classDef serving fill:#3D5550,stroke:#6BB7AA,stroke-width:2px,rx:8,ry:8,color:#E0E6ED classDef governance fill:#5A3F52,stroke:#C28BB0,stroke-width:2px,rx:8,ry:8,color:#E0E6ED USER[User Query] -->|Route| SUPERVISOR[Supervisor Agent] SUPERVISOR -->|Delegate| SQL_AGENT[SQL Agent] SUPERVISOR -->|Delegate| DOC_AGENT[Documentation Agent] SUPERVISOR -->|Delegate| CODE_AGENT[Code Agent] SQL_AGENT -->|Query| DB[Unity Catalog Tables] DOC_AGENT -->|Search| VS[Vector Search] CODE_AGENT -->|Execute| NB[Notebook Execution] SQL_AGENT -->|Return| SUPERVISOR DOC_AGENT -->|Return| SUPERVISOR CODE_AGENT -->|Return| SUPERVISOR SUPERVISOR -->|Synthesise| RESPONSE[Final Response] USER:::source SUPERVISOR:::governance SQL_AGENT:::processing DOC_AGENT:::storage CODE_AGENT:::serving DB:::ingestion VS:::storage NB:::ingestion RESPONSE:::source

    *Multi-agent orchestration: a supervisor routes queries to specialised agents, gathers results, and synthesises a unified response.*

    %%{init: {"theme":"base","themeVariables":{"background":"#0B0E14","primaryTextColor":"#E0E6ED","lineColor":"#5D6470","darkMode":true,"primaryColor":"#2E4A4A","secondaryColor":"#374151","secondaryTextColor":"#E0E6ED","tertiaryColor":"#111827","tertiaryTextColor":"#E0E6ED","edgeLabelBackground":"#1f2937"}}}%% graph TD classDef source fill:#3F4B59,stroke:#9CA3AF,stroke-width:2px,rx:8,ry:8,color:#E0E6ED classDef ingestion fill:#5A4B36,stroke:#C9A86B,stroke-width:2px,rx:8,ry:8,color:#E0E6ED classDef processing fill:#535072,stroke:#8E82B4,stroke-width:2px,rx:8,ry:8,color:#E0E6ED classDef storage fill:#2E4A4A,stroke:#5FAFA8,stroke-width:2px,rx:8,ry:8,color:#E0E6ED classDef serving fill:#3D5550,stroke:#6BB7AA,stroke-width:2px,rx:8,ry:8,color:#E0E6ED classDef governance fill:#5A3F52,stroke:#C28BB0,stroke-width:2px,rx:8,ry:8,color:#E0E6ED PATTERNS[Orchestration Patterns] --> SUPERVISOR_P[Supervisor Pattern] PATTERNS --> SEQUENTIAL[Sequential Pipeline] PATTERNS --> PARALLEL[Parallel Fan-Out] PATTERNS --> HIERARCHICAL[Hierarchical Delegation] SUPERVISOR_P --> ROUTE[Route by Intent] SUPERVISOR_P --> SYNTH[Synthesise Results] SEQUENTIAL --> STEP1[Agent 1 - Analyse] SEQUENTIAL --> STEP2[Agent 2 - Plan] SEQUENTIAL --> STEP3[Agent 3 - Execute] PARALLEL --> FAN[Fan-Out Queries] PARALLEL --> MERGE[Merge Results] PATTERNS:::governance SUPERVISOR_P:::processing SEQUENTIAL:::storage PARALLEL:::serving HIERARCHICAL:::ingestion ROUTE:::source SYNTH:::source STEP1:::source STEP2:::source STEP3:::source FAN:::source MERGE:::source

    *Multi-agent orchestration patterns: supervisor, sequential, parallel, and hierarchical.*

    Key Terms

    Prerequisites and Setup

    • Databricks workspace with Agent Framework, Model Serving, and Vector Search.
    • Multiple serving endpoints or tool sets for each specialised agent.
    • databricks-agents package.

    Step-by-Step Implementation

      Configuration Reference

      Multi-Agent Systems: Orchestrating Agents at Scale configuration options
      ParameterDefaultDescription
      max_delegation_depth3Maximum nesting depth for agent delegation
      timeout_seconds60Maximum time per agent execution
      parallel_executionfalseRun fan-out agents in parallel
      synthesis_modelsame as supervisorModel used for result synthesis

      Monitoring, Cost, and Security Considerations

      Common Pitfalls and Recommended Patterns

        Frequently Asked Questions