Rerankers
A reranker re-orders your search results for relevance. Search has to be fast, so it casts a wide net and returns the likely matches quickly. A reranker then takes the top handful, reads each one against the query with a heavier, more careful model, and puts the best ones first. It is the difference between “these all contain your words” and “this is the one you actually meant” — using an AI model you bring, with your own API key.
Auto-plays · use Back / Next to step through at your own pace.
Add a reranker
- Go to Models → Rerankers and click Add Reranker.
- Pick a provider — Voyage, Cohere or Microsoft Foundry. Reranking runs on your own key: your models, your limits, your bill.
- Choose the model and paste your provider API key.
- Click Add. Your key is verified with a live call before the reranker is saved, and the connection is re-checked regularly afterwards — the health dot next to the model shows any problem first.
Attach it to a list or group
A reranker does nothing on its own — you attach it to a list or a group, and from then on every search there is reranked before the results come back. The reorder happens inline, in the same request, so your front-end needs no changes. Attach the same reranker to as many lists as you like.
Reranker or judge?
Both use an AI model to improve results, but they do different jobs — and they compose well:
| Reranker | Judge | |
|---|---|---|
| Job | Re-orders results by relevance | Keeps or drops results against a rule |
| When | Live, inline with the search | After the search, in the background |
| Result | Same results, better order | Fewer results, all clearing your bar |
A common pairing: a reranker sharpens the order, then a judge filters out anything that still doesn’t meet a quality bar you set in words.
A worked example
Say a shopper searches your store for “warm waterproof jacket for winter hiking”. Plain search returns everything mentioning those words — including a summer rain shell and a heavy coat with no waterproofing. A reranker reads the top results against the full phrase and pushes the genuinely warm, waterproof, hiking-suitable jackets to the top, where shoppers actually look. Same catalogue, same query — a noticeably better first screen.