{"collection":{"slug":"browser-automation","name":"Browser Automation & Scraping"},"digest":{"id":"cwd_qB_RVRgsbqULqUPcIu04a","weekStart":"2026-05-11","title":"Firecrawl parses documents, Stagehand slims context, replays go live","intro":"Firecrawl's v2.10 adds a document parsing endpoint and lockdown mode, Browserbase ships embeddable session replay streaming, and Stagehand 3.4 introduces `ignoreSelectors` to trim agent context bloat.","body":"### Firecrawl opens up document parsing and tightens security\n\nThe biggest story of the week is [Firecrawl v2.10](/release/rel_hdj2Y_v5zDv3tGkS-px9g-firecrawl-v2-10), which lands with a new `/parse` endpoint that lets you upload local files — PDF, DOCX, XLSX, HTML, and more — up to 50 MB and receive clean, LLM-ready Markdown, JSON, or a summary. The parsing engine is written in Rust and claims to be up to 5x faster than previous approaches. Tables and reading order are preserved, and enterprise plans get full Zero Data Retention support. The endpoint is available across all of Firecrawl's SDKs, and the release also adds four new official SDKs for Go, Ruby, PHP, and .NET.\n\nAlongside parsing, the release introduces Lockdown Mode: setting `lockdown: true` on `/scrape` forces results to come exclusively from Firecrawl's index with zero outbound requests. It's a straightforward but powerful toggle for teams that need guaranteed no-crawl behavior. The release rounds out with new Question and Highlights output formats and a long list of reliability and security fixes.\n\n### Browserbase streams session replays to your own product\n\n[Browserbase session replay streaming](/release/rel_KrARqks2dPd_syuo5trIe-session-replay-streaming) is a notable addition for anyone embedding browser automation into a product. Session replays can now be consumed via an HLS streaming API — your backend fetches a `.m3u8` playlist, your frontend hands it to any HLS-compatible player, and end users stream segments directly from Browserbase's CDN. Playback is available within seconds of a session ending. The feature is free on every plan, up to 120 sessions per month.\n\n### Stagehand 3.4 cuts context noise and adds new model support\n\n[Stagehand 3.4.0](/release/rel_TKnMCgBz3U99gfRNFMGpf-browserbasehq-stagehand-3-4-0) (also released as stagehand/server-v3 v3.6.9) addresses a common pain point in agent-based browser automation: context bloat from noisy page elements. The new `ignoreSelectors` parameter on `extract()` and `observe()` lets you exclude ads, navigation, modals, related posts, and anything else your agent should skip. The release also makes agent variables available in the Stagehand API schema without the experimental flag, switches the default agent mode to hybrid for compatible models, and adds support for several new CUA models including `openai/gpt-5.4-mini`, `openai/gpt-5.5`, and `anthropic/claude-haiku-4-5`.","releaseIds":["rel_hdj2Y_v5zDv3tGkS-px9g","rel_KrARqks2dPd_syuo5trIe","rel_TKnMCgBz3U99gfRNFMGpf"],"releaseCount":7,"generatedAt":"2026-07-11T16:37:20.312Z","releases":[{"id":"rel_hdj2Y_v5zDv3tGkS-px9g","title":"Firecrawl v2.10","path":"/release/rel_hdj2Y_v5zDv3tGkS-px9g-firecrawl-v2-10","org":{"slug":"firecrawl","name":"Firecrawl"},"importance":null},{"id":"rel_KrARqks2dPd_syuo5trIe","title":"Session replay streaming","path":"/release/rel_KrARqks2dPd_syuo5trIe-session-replay-streaming","org":{"slug":"browserbase","name":"Browserbase"},"importance":null},{"id":"rel_TKnMCgBz3U99gfRNFMGpf","title":"@browserbasehq/stagehand@3.4.0","path":"/release/rel_TKnMCgBz3U99gfRNFMGpf-browserbasehq-stagehand-3-4-0","org":{"slug":"browserbase","name":"Browserbase"},"importance":null}]}}