Interface: GuardrailOutputPayload
Defined in: packages/agentos/src/safety/guardrails/IGuardrailService.ts:196
Payload for output guardrail evaluation.
Provided to IGuardrailService.evaluateOutput before response chunks are emitted to the client. Use this to filter, redact, or block agent output.
Remarks
Every guardrail is called for each chunk of the turn's output stream that
carries isFinal: true (the FINAL_RESPONSE, an ERROR the turn yields). With
GuardrailConfig.evaluateStreamingChunks set, it is also called for
each TEXT_DELTA chunk (real-time filtering), and with
GuardrailConfig.evaluateToolCalls set, for each TOOL_CALL_REQUEST
chunk. Other chunks are not evaluated, and neither is the ERROR that
processRequest() yields from its own catch or for an input guardrail's
BLOCK.
Properties
chunk
chunk:
AgentOSResponse
Defined in: packages/agentos/src/safety/guardrails/IGuardrailService.ts:201
The response chunk to evaluate
context
context:
GuardrailContext
Defined in: packages/agentos/src/safety/guardrails/IGuardrailService.ts:198
Conversational context for policy decisions
ragSources?
optionalragSources:RagRetrievedChunk[]
Defined in: packages/agentos/src/safety/guardrails/IGuardrailService.ts:209
RAG source chunks retrieved for this request. Available to output guardrails for grounding verification. Persists across all chunks in a stream (not just the final chunk). Undefined when no RAG retrieval was performed.