AI search, managed
The easy button for AI search.
Put data in → amazing agentic search out.
Evaluating Relevan for a large team? Talk to sales about dedicated capacity and an SLA.
01 / Drop it in
Your records become a tuned index.
02 / It learns
Agents recover on the second try, not the fifth.
03 / It improves
Every search makes the next one land closer.
04 / No ops
Nothing to size, tune, or page you at 3am.
01 / Ingest
Drop your records in.
Send your catalog, your docs, or your profiles. The index designs itself.
02 / Feedback
Your agents stop guessing.
Every response tells the agent what to fix, so it recovers on the second try instead of the fifth.
From a real response
Matched: headphones. Top results share terms you have not used yet: wireless, audio, noise, cancelling.
The agent reports back which result the outcome landed on, as one query id and one object id. That call is the whole signal.
03 / The loop
It gets better while you sleep.
Agentic search does not learn. Everywhere else an agent's search leaves no trace, so the next one starts from zero. Relevan keeps every search, every miss, and every outcome, and folds them back into the ranking.
noise-cancelling headphones for travel
Run the search to compare regular search with the Relevan Protocol.
This walkthrough is scripted against a fixed product catalog, so it plays the same way every time. Every query, score, and protocol response is replayed from the real search engine, and a test fails if any of them drifts. Create an index in the console to run it against your own data.
Searches to reach the answer
3 → 2
Results the agent must read
22 → 6
One shopper need, one product catalog, the same opening query. Run 1 keeps widening the query and hands back 22 results after three searches. Run 2 reads what matched, drops the term that hit nothing, and finishes at six. Source: the scripted demo episode in packages/demo.
04 / Ops
Nothing to operate.
Search engines are easy to start and hard to keep. Relevan carries the indexing, the tuning, the upgrades, and the uptime.
- —No cluster to size, shard, or upgrade
- —No analyzer or mapping to hand-tune
- —No relevance engineer to hire
- —No reindex window, and no pager
The whole surface
Two calls.
RELEVAN_URL="https://api.dev.relevan.dev"
# Search the index
curl -X POST "$RELEVAN_URL/v1/$ORG_ID/indexes/products/search" \
-H "x-api-key: $RELEVAN_API_KEY" \
-H "content-type: application/json" \
-d '{"query": "travel headphones"}'
# Report the outcome. Send back the queryId
# from the response's searchContext.
curl -X POST "$RELEVAN_URL/v1-beta/$ORG_ID/indexes/products/feedback" \
-H "x-api-key: $RELEVAN_API_KEY" \
-H "content-type: application/json" \
-d '{"actionName": "conversion", "label": "purchase",
"queryId": "<queryId>", "objectId": "sku-12345"}'The second call is the loop. Everything after it is ours.