Orchestrator LLM

An Orchestrator LLM is a top-level AI model that directs several sub-models to work together toward a single goal. Think of it like a team lead at a company who assigns tasks to team members and then pulls together their individual outputs into one final report.

For example, answering a single question well might require several models that are each good at different things—one skilled at search, one at writing code, one at summarizing documents. In this setup, the Orchestrator LLM breaks the question into pieces, decides which sub-model should handle which part, and then takes the results from each model and stitches them together into a final answer.

This kind of structure doesn't need multiple servers—it can run entirely on a single computer. In fact, in the local-execution version released by Perplexity, both the model acting as conductor and the sub-models it directs all run together on a single user device.

The article explains it as "a structure where the Orchestrator LLM, sub-agent LLMs, and the entire agent harness all run within the user's hardware." It's worth noting that Orchestrator LLM isn't a newly invented, standalone product—it's a term for the model that plays the conductor role when multiple AI models are linked together into a single agent system.