Structs§
- Agent
Inference Request - Compaction
Progress - Completion
Metadata - Hosted
Tool Call Completed - Hosted
Tool Call Started - Hosted
WebSearch Config - Inference
Capabilities - Inference
Failure - Inference
Provider Context - Inference
Provider Metadata - Inference
Turn Context - Instruction
Bundle - Message
Delta - Model
Descriptor - Model
Harness Profile - Model
Profile Reasoning - Model
Selection - Output
Config - Reasoning
Config - Reasoning
Delta - Reasoning
Effort Descriptor - Runtime
Hints - Speed
Policy Decision - Token
Usage - Tool
Call Completed - Tool
Call Delta - Tool
Call Started - Tool
Search Config - Tool
Search Config Overlay - Turn
Tool Outcome - Result of executing one tool call via
TurnToolExecutor.
Enums§
- Effective
Tool Search Mode - Hosted
WebSearch Mode - Inference
Event - Model
Instruction Overlay - Model
Schema Policy - Provider
Auth Type - Provider
Family - Runtime
Profile - Speed
Policy Phase - Tool
Search Mode - Tool
Search Mode Error - Tool
Search Provider Variant
Traits§
- Inference
Engine - Turn
Tool Executor - Executes a single tool call through the runtime’s registry + policy. Implemented by the runtime; used by providers that run their own in-stream agent loop.
Functions§
- cache_
hit_ rate - finish_
reason_ from_ stop_ reason - Canonical mapping from provider-native stop reasons to the finish reason
surfaced as
finishReasononturn/completed. Only the terminal inference step’s stop reason reaches the turn surface, sotoolUseappears only when a turn genuinely ends on a tool-use step (e.g. tool rounds exhausted). Unknown stop reasons pass through unchanged. - provider_
supports_ streaming - provider_
supports_ tools