# answer_documents A RocketRide adapter node that turns model/agent answers into embeddable documents, so validated, self-describing facts can be indexed back into a vector store. ## What it does Reads the `answers` lane and emits documents on the `documents` lane, bridging a gap in the pipeline graph: many nodes produce `answers`, but none previously turned an answer back into `documents`, so an LLM's structured/validated output could not be re-embedded and indexed. Granularity is "one document per validated fact": - a JSON array answer yields one document per item (each fact is indexed independently), - a JSON object or plain-text answer yields a single document, - an empty/whitespace answer yields no documents. Each output document's `page_content` is the fact text (list items that are already strings are used verbatim; other values are serialized as JSON). Documents are stamped with an incrementing `chunkId` (reset for each input object) and marked as non-table content (`isTable: false`, `tableId: 0`). The original answer is always passed through unchanged on the `answers` lane, so a single pipeline can both index the answer and keep serving it (for example to a response node) in the same run. The node has no Python package requirements of its own: it relies entirely on the separately installed AI module. --- ## Configuration ### Lanes | Lane in | Lane out | Description | |-----------|------------------------|-------------------------------------------------------------------------| | `answers` | `documents`, `answers` | Emit one document per validated fact; pass the original answer through | No configuration fields. Wire answers in. --- ## Usage Place this node after an LLM, agent, or extract/validate node (anything that produces `answers`) and before an embedding node and vector store: ```text llm / agent → answer_documents → embedding_* → vector store ↘ response (original answer, optional) ``` The `documents` output is embedded and indexed for later retrieval, enabling the highest-quality RAG design — indexing self-describing, validated facts rather than raw chunks. The `answers` passthrough lets the same answer continue to a downstream sink in the same pipeline. Example: an upstream node returns a validated JSON array of facts ```json [ {"metric": "revenue", "period": "FY2024", "value": "5.2M", "currency": "USD"}, {"metric": "gross_margin", "period": "FY2024", "value": "61%"} ] ``` `answer_documents` emits two documents, one per fact, each ready to be embedded and indexed. --- ## Schema _No configuration fields._ ## Source [ View source](https://github.com/rocketride-org/rocketride-server/tree/develop/nodes/src/nodes/answer_documents)