We benchmark against 10 other search APIs on fresh news queries. The median was around 1.2s; we came in around 166ms and scored highest on answer accuracy (89% vs 84% for the next cluster).
Latency matters because agents loop. A 1.2s first call eats the budget for follow-ups — you get one shot at framing the query. At sub-250ms the agent can actually search, read, reformulate, and search again.
Measuring this stuff carefully is something I've been at for a while. My ECIR 2019 paper (linked below) was an exhaustive study of 11 index compression methods across 5 query processing algorithms on standard collections — the codebase became PISA, which a lot of IR folks still use for research. Almost ten years later, the workload has changed completely (agents, not humans), but the benchmarking discipline is the same.
ECIR 2019 paper: https://www.antoniomallia.it/uploads/ECIR19c.pdf
Pisa Engine: https://github.com/pisa-engine/pisa
Full methodology and charts for Seltz: https://seltz.ai/blog/why-we-built-seltz
https://console.seltz.ai/login
Hi HN, Antonio here. Founder of Seltz.Seltz is a web search API built for AI agents. We wrote the crawler, the index, and the retrieval models ourselves, in Rust, by a team that's spent years building web search at scale. In our tests, queries come back in under 200ms.Efficiency was the first design principle. Search sits on the critical path: agents can't generate their first tokens or kick...
https://github.com/pisa-engine/kentro
Double-digit speed-ups over naïve K-Means Parallel by default (Rayon) Balanced & spherical variants in one crate Memory-savvy for multi-GB datasets Builder API that just feels ergonomicDocs are live, the license is Apache-2.0, and the repo is one `cargo add kentro` away. Give it a spin—curious to hear how it performs in your pipelines!
https://github.com/pisa-engine/BMP
We built BMP, a fast and memory-efficient search engine for learned sparse retrieval — written in Rust and with Python bindings.It supports exhaustive (non-approximate) search over large collections like MS MARCO, without dropping query terms or pruning the index.Features:- Full support for SPLADE, uniCOIL, CSV, and similar models- No static pruning – keeps full index fidelity- No term dropping...
Also, think about `Windows` or `Go`, there is no way you get confused once these words are in the right context.