Crate llmweb 

llmweb

Extract structured data from any webpage by combining a headless browser with an LLM.

• 5 preprocessing modes: Html (cleaned), RawHtml, Markdown, Text, Image.
• Code generation (route A): the LLM emits a JS extractor that runs in the browser via tab.evaluate — store it, replay it without further LLM cost.
• Selector recipe (route B): the LLM emits a declarative CSS-selector recipe, executed in pure Rust against any HTML.

Re-exports

pub use error::LlmWebError;

pub use error::Result as LlmWebResult;

pub use preprocess::Format;

pub use preprocess::Preprocessed;

pub use preprocess::RunOptions;

pub use recipe::ExtractRecipe;

pub use recipe::FieldRule;

pub use streaming::PartialStream;

Modules

error

openai

Re-exports from async-openai so users can build a custom client without taking a direct dependency on the crate version.

preprocess

Page preprocessing.

recipe

Route B — declarative selector recipes.

streaming

Incremental JSON streaming.

Macros

strip_markdown_backticks

Structs

Browser

Thin wrapper around a headless_chrome::Browser configured with stealth flags.

LlmWeb

The main client. Holds an async-openai client and the model name.

Tab

A handle to a single page. Exposes methods for simulating user actions (clicking, typing), and also for getting information about the DOM and other parts of the page.

Functions

run_recipe_on_tab

Apply a recipe against the current state of a tab the caller has already navigated. No LLM call. Uses raw HTML so attributes (href, src, etc.) the recipe depends on are preserved.

run_script_on_tab

Run a previously-generated JS extractor against a tab the caller has already navigated. No LLM call.