Every meaningful change to the public API or product. Chronological, dated, honest.
/v1/cloud/proxy-fetchPOST /v1/cloud/proxy-fetch — bytes-billed HTML scrape via spider.cloud proxy. 4 pools (datacenter, isp, residential, mobile), 199 countries. ~13× cheaper than /v1/cloud/scrape for plain HTML./v1/cloud/scrape gains force_backend: "proxy" for cheap fetches./v1/actors/check-links gains cloud_retry_mode: "proxy" — 30× cheaper, 12× faster retry path./v1/crawls accepts proxy_pool + proxy_country top-level. Server resolves credentials; customer never sees a vendor key.sitemap-audit default max_urls lowered 1000 → 100 (was a silent billing landmine — 200 URLs probed = 200 page-credits per call).sitemap-audit dry_run: true — discovers sitemaps + counts URLs without probing or billing.sitemap-audit response gains pages_billed as an explicit billing signal./v1/cloud/balance USD display was off by 10× (showed $1.55 instead of $15.46). Per-call cost values were already correct.FOR UPDATE SKIP LOCKED. Safe to run on 3+ processes against one Postgres primary.crawl_runs.worker_id retained on completion so "which worker handled this run" is answerable post-mortem./v1/actors/render-diff — AI-bot blind %. Compares static HTML to JS-rendered DOM. The 2026 AEO metric./v1/actors/internal-link-graph — PageRank + WCC + orphan detection on any existing crawl_id. 1 flat credit per call./v1/actors/sitemap-audit — 7-bucket sitemap health (ok / redirect / 4xx / 5xx / noindex / canonicalised away / network error).check-links + CF email-protection filtercloud_retry: true reclassifies broken links via chrome render. Rescues LinkedIn 999, Cloudflare anti-bot, paywalled news.mailto: in the browser)./v1/actors/structured-data — JSON-LD, Microdata, RDFa, OpenGraph, Dublin Core, Microformats in one call. Backed by extruct./v1/actors/audit-onpage — ~30 on-page SEO rules per call./v1/actors/check-links — Lychee-based broken-link check, up to 200 URLs per call./v1/actors/extract-article — Trafilatura body extraction with author + date metadata./v1/cloud/screenshot validates returned bytes start with a known image magic header (PNG, JPEG). Catches vendor-side response corruption.WORKER_PARALLEL_JOBS default 4 → 8; tier-aware per-request concurrency caps (internal = 64).pages + cloud_pages. No multipliers, no credit dust.Want the engineering-level detail? Daily session logs live in vikasswaminh/crawlcrawl/docs.