[hackgician]

Yeah, I think someone opened a similar issue on GitHub: https://github.com/browserbase/stagehand/issues/389

Repeatability of extract() is definitely super interesting and something we're looking into

[9dev]

Cache the response for a given query-page hash pair maybe? So the LLM will only be consulted when the page content hash changes, the previous answer be reused otherwise