Agent step labels
IQ labels each request with the kind of agent-loop step it represents, such as planning, a tool call or a summary, and shows the label on the request's detail, so you can see how an agent loop's work and spend break down step by step. Labels describe the work, not a model, so they stay stable as models change.
The step-class vocabulary
Each request gets one of six step labels:
plannerfor the decomposition step in an agent loop, where the agent breaks a goal into subgoals or lays out an ordered plan.tool-arg-formatterfor the step that emits a function call against a JSON schema, where the output shape is the structured argument blob.summariserfor the step that condenses a long input into a short output.extractorfor the step that pulls specific fields out of a larger body of text.coderfor the step that writes or reviews code.criticfor the step that evaluates a prior output against a specification.
The labels are mutually exclusive; each request gets the single label that best fits the step. When no label clearly applies, the request has no step label.
Thinking budget
Each request also carries a thinking budget of none, low, medium or high, which sets how much extended reasoning the model spends on it. IQ translates the budget to each provider's own setting (an Anthropic thinking budget, an OpenAI reasoning effort, or a Gemini thinking level), so your call site doesn't change. You can set it per request, rely on your organization's default, or let IQ set it; each request records which applied.
Seeing what served each step
The x-iq-served-model response header names the model that served each request, and the request's detail shows its step label, thinking budget and cost. Organizations can opt in to letting low-stakes steps run on a lower-cost model; steps whose output steers the rest of the loop are never moved down. Your organization's focus mode (cost, balanced or quality) applies to every step.
Latency and replay
Labeling adds no network round trip. Each label is saved with the request, so the session view can replay an agent loop step by step.
Related pages
Routing explains the routing modes you can choose. Capability aliases explain the cap: handles you can use as a model.