quillsift

Python bindings for QuillSift: a Rust-native, self-hosted scraping engine with a deterministic (LLM-free) path, a legacy bounded crawler and schema-validated LLM extraction.

QuillSift was previously published as scrapeowl. That package is superseded by quillsift; the API is otherwise the same.

pip install quillsift

import quillsift

markdown = quillsift.fetch("https://example.com")
data = quillsift.fetch("https://example.com", format="json")

schema = {
    "type": "object",
    "required": ["title"],
    "properties": {"title": {"type": "string"}},
}
result = quillsift.extract(
    "https://example.com",
    schema,
    model="gpt-4o-mini",  # reads OPENAI_API_KEY by default
)

report = quillsift.crawl(
    "https://example.com",
    max_depth=2,
    max_pages=25,
    respect_robots=True,
)
for page in report["pages"]:
    print(page["url"], page["content"]["metadata"]["title"])

fetch never needs an LLM or credentials. crawl is a legacy fallback: it is breadth-first, same-origin by default, deduplicates URLs and returns per-page failures without aborting the job, but new crawling and JavaScript-rendering work should use MudFish. extract supports provider="openai" (OpenAI-compatible /chat/completions) and provider="ollama" (/api/chat); output must parse as JSON and validate against the schema, with bounded repair.

Security defaults: private/loopback destinations are refused unless allow_private=True (needed for a local Ollama endpoint), redirects and response sizes are bounded, and API keys are read from environment variables or provider headers, never embedded in URLs.

Apache-2.0. Not affiliated with ScrapeGraphAI.