Component Grounded Generator

Grounded Generator Component

Provide the generator capability for an Agentic RAG slot.

Interface

Input GenerationRequest:

• query: non-empty original question
• documents: selected evidence documents. Each document must contain non-empty id and text; an empty sequence is allowed and is rendered as [No evidence supplied].
• optional max_tokens: forwarded to the frozen Executor Model

Output GenerationResult:

• answer: non-empty generated grounded answer. Leading and trailing whitespace from the model response is stripped.

Execution

Run scripts/component.py:run(inputs, context). The runtime must provide the frozen context.call_model(...). The component formats the supplied evidence, calls the model with temperature=0.0, and rejects malformed inputs or empty model responses. Do not retrieve documents, rerank documents, or modify model parameters.