Markdown for reading, annotated DOM for acting.
Scrape bifurcates each page into extracted Markdown plus a durable set-of-marks map for interactive elements.
Published alpha - @cuitty/scrape
Paste a goal. Get data and a scraper.
Cuitty Scrape turns a URL plus plain-English intent into extracted results, a traceable run timeline, and a reusable artifact that can replay without a model.
43offline tests passing
P0-P3engine and integrations shipped
1 cmd
cui scrape testBuilt for agents that need browsers, state, and proof.
Compare capabilitiesEvery useful run becomes an artifact.
Click and fill steps are emitted as a reusable scraper. Replay runs without a model and heals selectors on miss.
CAPTCHA, login, and destructive actions pause safely.
The server emits interaction-required events and resumes after a client or operator resolves the blocker.
Cuitty Safe handles secrets and sessions.
Credential refs resolve outside the scrape package, and browser storage state can be captured for reuse.
Use it as a package, HTTP service, or Pilot executor.
Frontend clients, backend server, CLI, CUI runner, Pilot delegation, and DSL adapter share one typed contract.
Mocks are default; heavy deps are lazy.
The package builds without Playwright, Anthropic, or Cuitty Safe installed. Real implementations are opt-in.
Docs for the package, the protocol, and the suite.
Quickstart
Install, start the mock server, run one scrape, and inspect the artifact.
Client guide
Use createScrapeClient with polling, events, and interactions.
HTTP API
Routes for jobs, runs, events, views, artifacts, replay, and health.
Capabilities
Feature matrix for engine, protocol, auth, UI surfaces, and publishing.