Key Info
Perplexity introduced Fast Search in its Search API, running on Photon, a Rust-based retrieval and ranking service built by a small team with hundreds of agents, returning 95% of search results in 230 ms or less.
Highlights
- Photon builds indexes separately from live query servers, updating one serving group at a time with cache warming and readiness checks while others keep serving.
- Inside Photon, p99 response time dropped from ~800 ms to ~65 ms, using ~20% fewer serving machines and storing 2.5x more data per document.
- Links to the engineering blog and Fast Search getting-started guide were shared.
