Perplexity Adds Fast Search API on Rust-Based Photon Engine
- Source
- x.com
- Date
Search calls are one of the most repeated operations in an agent loop, so their latency and price compound quickly. Perplexity's new Fast Search API targets that directly. It runs on Photon, a retrieval engine rebuilt in Rust, and Perplexity reports 95% of results returned in under 230 milliseconds. The company puts the saving at 68% lower cost per agentic task compared with its default search preset. Perplexity is candid that the speed comes with a modest relevance tradeoff, so this is a tier choice rather than a free upgrade. For builders, the sensible pattern is to route routine lookups, such as checking a fact or fetching a known page, to the fast tier and keep the default preset for queries where recall matters more than speed. Separately, Perplexity said its Portable Computer can now run its agent harness and model fully on AMD Ryzen AI Max PCs, calling the cloud only with permission.
Introducing Fast Search in the Perplexity Search API. Fast Search runs on Photon, our new Rust-based retrieval and ranking service that we built with a small team of engineers and hundreds of agents. It returns 95% of search results in 230 ms or less.
Context
Perplexity has added a speed-focused mode to its Search API, the service that lets other software pull Perplexity's web results; Nous Research's Hermes , for example, can call it. Perplexity says the new Fast Search returns 95% of search results in 230 milliseconds or less. That kind of figure describes the slow cases rather than the average: by Perplexity's account, only the slowest 5% take longer than 230 milliseconds, which matters for an agent that may run many searches before it answers.
Fast Search runs on Photon, which Perplexity describes as a new retrieval and ranking service written in Rust. Retrieval finds candidate pages for a query and ranking decides their order, the two steps that determine both what comes back and how long it takes. Perplexity says a small team of engineers built Photon with hundreds of AI agents. The post does not say what work the agents did or how their contribution was measured.
Speed is the only thing the post's text measures. The existing Search API had already been tested by Artificial Analysis across three context settings that control how much extracted content each result carries, and Perplexity said those settings took the top three spots on that search index. Whether Fast Search keeps that result quality at lower latency is not answered by this announcement.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Checking sign-in…
Loading comments…