Agent Studio latency lab

See the request travel
before you blame the model.

This small benchmark separates the time spent reaching Algolia Search from the time spent getting an Agent Studio response. It is a teaching demo, not a production dashboard.

Controlled runSame query setPlain-English readout
What we are comparing

Three paths, one question

01
Raw Search

How long does a direct search request take?

02
Agent Studio · fixed index

Search plus agent reasoning with the index already configured.

03
Agent Studio · request-selected index

Search plus agent reasoning when the application supplies the index for this request.

i

What the numbers mean: p50 is the middle result. p95 is a slower result that only about one request in twenty exceeds. Looking at both helps us avoid being fooled by one unusually fast or slow request.

Run the CLI in the browser

Start a controlled run

Demo mode

The browser will run the same CLI with deterministic synthetic responses. No credentials are needed.

3 queries · 1 warm-up · 3 measured runs per case

agent-studio-latency-benchmarkReady
$ npm run benchmark -- --progress
Press “Run benchmark” to start the controlled comparison.