Now legacy, this Agent Bricks offering coordinates agents and tools from different domains, and every end user must be able to reach each subagent personally. Genie Agents replace it for new builds, with the Agent Bricks CLI as the code route.
Also called Multiagent Supervisor, Multiagent Supervisor.
Read more: Databricks docs
In the Ultra Transcenders books
Each book explains Supervisor Agent in context, with comparison tables and the common traps.
Terms in this definition
- Legacy
A label Databricks gives to older agent offerings: their documentation remains so that existing agents can be kept running or migrated, but new agents should avoid them.
- Agent Bricks
Code-first tooling from Databricks for creating and shipping agents, including AgentKit, a CLI, a runtime, and memory and session features. Low-code agents are a separate product, Genie Agents.
- Chat message roles
Labels on chat messages: instructions go under system, the person's input under user, the model's previous answers under assistant, and results returned by a called tool under tool (or function).
- Genie
Genie Agents (once called Genie spaces) let Azure Databricks users ask about curated data in everyday language; Genie writes a read-only SQL query behind the scenes and shows what it finds, with charts.
- Agent Bricks CLI
The
agentbrickstool, in Beta and installed through thedatabricks-agentbrickspackage, creates the skeleton of a custom agent, lets you try it on your own machine and then ships it to Agent Runtime.
Related terms
- long_task
Lets a Supervisor Agent avoid a time-out: set in
databricks_options, it makes the supervisor stop in time and sendtask_continue_request, after which the application's next message resumes the work. - Web search
Answers backed by up-to-date pages from the web. Gemini and OpenAI foundation models have their own native search,
system.ai.web_searchprovides it as a managed MCP service, and Supervisor Agent seeks the user's approval before each search.