Riven Grounding

Retrieval and citation: decides when an answer needs sources, finds them, and binds each claim to something a reader can check.

Grounding is what turns a plausible answer into a checkable one. It runs between the question and the model: retrieve relevant passages, put them in front of the model, and attach the citations to the claims they support.

When it engages#

Not every question needs the web. Asking to rewrite a paragraph does not; asking what happened this week does. Grounding classifies the request and only pays the retrieval cost when currency or verifiability is actually at stake — which is why some answers carry citations and others do not.

How retrieval works#

1. Rewrite the user's question into retrieval queries, using thread context so follow-ups do not need restating.
2. Search the Web Index both semantically and lexically, then merge the result sets.
3. Rerank passages for relevance and source quality, and drop near-duplicates so one story does not crowd out the rest.
4. Assemble a bounded evidence set that fits the context budget, preferring diverse sources over many passages from one page.
5. Generate the answer against that evidence, binding each citation to the passage it came from.

Citation integrity#

A citation points at a passage the platform actually retrieved, not at a URL the model produced from memory. If retrieval turns up nothing usable, the honest outcome is an answer that says so — a fabricated source is a worse failure than an admitted gap.

Using it yourself#

Research-tier models reach grounding directly: pin riven-research (or a Perplexity sonar model) and answers come back web-grounded with sources. For a fully custom pipeline, embed your own corpus with riven-embed and supply passages yourself — see API reference → Embeddings.

curl https://api.rivenai.io/v1/chat/completions \
  -H "Authorization: Bearer $RIVEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "riven-research",
    "messages": [{"role": "user", "content": "What changed in EU AI regulation this month?"}]
  }'