OpenRouter's Advisor tool lets the model handling a task ask another model for guidance and then continue its own work. The consultation runs on the server and returns text as a tool result. Developers can choose models from different providers for the two roles.
This supports a workload design in which a cheaper model handles routine steps and consults a more expensive one selectively. The advisor gives a single-turn answer without running its own agentic loopThe cycle an agent runs in: decide, call a tool, read the result, decide again — repeating until the goal is met or a stop condition fires.Full definition →, and each model's usage is billed separately. The announcement establishes that mechanism; its illustrative pricing scenarios are not measured savings for a deployed workload.
Terms in this piece · Glossary
agentic loop — The cycle an agent runs in: decide, call a tool, read the result, decide again — repeating until the goal is met or a stop condition fires.