Agent classes / RAG agent / Rerank retrieved chunks for RAG
Rerank retrieved chunks for RAG for a rag agent
Updated 2026-10-27 (newest stored date on this page) · score asOf 2026-09-14 (stale)
The class: Answers from a retrieved corpus and says when the corpus does not support the claim.
- Passage ids on every factual sentence
Depot Pick: Voyage AI
When not to use it: Already on Cohere embed+rerank in one bill — stay; want EU all-in-one embed/rerank/Reader — Jina; tiny corpus where embeddings alone suffice — skip rerank.
Runner-up: Cohere.
Cheapest matching sourced SKU (usd_per_million_tokens): Voyage AI rerank-3-lite $0.02 — source, asOf 2026-10-03. All sourced prices
Failure modes this class and job share (7)
- Stale retrieval: Store the source URL and the time it was fetched. If a newer canonical page exists, or two stored sources disagree, do not answer from the older one.
- Context overflow: If the input exceeds the window, say so and retrieve a narrower slice. Do not answer from a truncated head as if it were the whole source.
- Eval contamination: Keep gold labels out of the agent prompt. Store the harness version beside the score. Do not compare a number across a benchmark revision without saying the revision changed.
- Lost in the middle: Ask the agent to name the passage it used. Spot-check questions whose only support sits in the middle of the context, not only at the edges.
- Policy conflict: If cited sources contradict each other, return the conflict and the URLs. Do not pick a winner silently.
- Silent truncation: Tools must return whether the payload was truncated and how long the original was. Fail the task if the answer depends on the missing span.
- Ungrounded answer: Require a source URL or passage id on factual sentences. Drop or mark any sentence whose source does not contain the claim.