Model provider request
One round trip in which a harness sends context to a provider and receives model output.
Illustrative conversation
- Why did one assistant reply create four provider requests?
- The agent made a model decision, used a tool, inspected the result, and then made another request to finish the turn.
Definition
A model provider request is one invocation of a hosted or local model interface. The harness selects a model and settings, sends the current input, and receives output plus metadata such as status and token usage. One user-visible turn can contain several provider requests when an agent alternates between model decisions, tool execution, and verification.
Reviewed