Get started · Evaluate
Evaluate us
Run a head-to-head comparison against your current retrieval provider in an afternoon.
Pick a benchmark
Use 50–200 representative queries from your real workload. Synthetic test sets miss the long tail; real query logs surface freshness and ranking gaps.
What to measure
- Recall@10 — does the right answer appear in the top results?
- Freshness — for time-sensitive queries, how recent is the top result?
- Citation accuracy — when used in a RAG pipeline, do generated answers cite real sources?
- p95 latency under your typical concurrency.
Sample evaluation script
The Examples section ships an end-to-end eval harness for Search and Research that you can adapt.