v0.36.0

0.36.0 (2026-09-19)

Features

• scrape: cache the fetch so a repeat scrape skips the network (601d5f5)

Bug Fixes

• crawl: do not fail a job because robots.txt could not be read (e8eea02)
• crawl: keep the last robots rule of a file that lands on the size cap (1e41fb0)
• extract: accumulate usage across legs, and report it on /v2/search (6623dce)
• extract: forward reasoning_effort from the structured builder (dc161c1)
• extract: stop corrupting markdown for non-HTML sources (0ae7c40)
• extract: stop the rescue candidate from reintroducing page chrome (8272600)
• harden consolidated crawl reliability changes (5691492)
• mcp: let CRW_LOCAL actually keep the run local (abc05e2), closes #557
• preserve crawl and renderer reliability invariants (65f385a)
• renderer: connect CDP over every resolved address, not just the first (77c0c74)
• scrape: scope the page cache per tenant (e0a583b)
• sdk: return the map sitemaps and keep local mode local (76c59c0), closes #557
• v2: carry llmUsage through to_v2_document (85fe020)