Browserbase describes using TypeSafe's Jev, a small model that returns a typed answer plus probabilities rather than free text, to choose browser actions inside Stagehand, Browserbase's framework for letting AI control a web browser. Jev first classifies an instruction, such as clicking or filling a field. Stagehand builds a list of possible targets, and Jev selects a target and assesses whether any candidate matches. Stagehand executes accepted choices and falls back to a LLMA large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.Full definition → otherwise. In Browserbase's early testing, median Act latency fell from 1.97 to 0.46 seconds. The classifier handles a bounded decision while the surrounding software supplies candidates, execution and fallback.
Terms in this piece · Glossary
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
