Intel / repo

firecrawl v2.11.0 — Firecrawl v2.11.0

Source: github.com
Author: rhys-firecrawl
Date: Published Jun 19, 2026

Context

Firecrawl v2.11.0 adds a way to extract structured data from repeated website scrapes without asking a to interpret every request. Its deterministicJson format generates an extractor for the requested data structure and caches it for that site.

Subsequent scrapes can reuse that extraction logic. The release describes this as reducing repeated processing costs and making results more consistent. The change separates creating the extraction method from applying it repeatedly; it does not establish that generating the method never uses an AI model.

Key takeaways · AI-distilled

Browser sessions now return the raw CDP (Chrome DevTools Protocol) WebSocket URL, so you can attach Playwright or Puppeteer to a live Firecrawl browser and drive it yourself instead of being limited to Firecrawl's own actions.
The monitor's LLM judge used to invent before/after text on long pages. The fix was evidentiary: give it the full unified diff as its only input, so it can only describe changes that actually appear.
JSON-mode monitor diffs now use order-insensitive equality. Arrays with identical values in a different order were previously reported as changes, firing an alert on every single run.
The research index bills as search: 2 credits per 10 results, or 10 credits per 10 for zero-data-retention teams. Keeping your data out of retention costs 5x on this endpoint.

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.

Key quotes

“It has state-of-the-art recall on arXivQA, outperforming the next best provider by 18% at comparable cost.”

“Added a `redactPII` option that strips personal and sensitive data like names, emails, phone numbers, addresses, and secrets out of scraped content before it's returned.”

“Firecrawl generates a reusable extractor for your schema and caches it per site, so repeat scrapes are cheaper and return consistent results.”

“Added a `goal` to monitors so an LLM judges each detected change as meaningful or noise against what you actually care about, cutting alert spam and surfacing the changes that matter first in summary emails.”

Read the source github.com