Firecrawl turns Alexandria into a data marketplace while Browserbase unlocks logged-in pages

Alexandria becomes the product

Firecrawl's week was less about crawling and more about what the crawl feeds. Three announcements all point the same direction: Alexandria, its web-data layer, is being packaged into ready-made datasets that chat assistants and coding agents can query directly.

The most consumer-facing is Directory, a free recruiting dataset for ChatGPT and Claude: describe a role, get a shortlist drawn from 250M+ profiles with experience and work emails attached. On the developer side, Firecrawl added cybersecurity data to Alexandria for dependency checks, letting Codex and Claude Code agents compare a project's dependencies against 400,000+ known vulnerabilities and patch them before anyone else finds them. And the distribution play: Firecrawl is now an integration in eve, Vercel's open-source agent framework, where a single command wires web access into an agent. Stacked together, these read as a shift from "we fetch pages" to "we sell the answers," with the integration doing the funnel work.

Fetch gets past the login wall

Browserbase's changelog entry is small in scope but big in practice. Fetch can now reuse saved login cookies from a Context by passing a contextId, returning pages behind a login as raw content, markdown, or structured JSON. The workflow is the one scrapers already know: log in through a browser session with persist: true, end the session to save cookies, then point Fetch at the Context. One caveat worth planning around — authenticated Fetch reads cookies from the last saved session and never writes updated ones back, so token refreshes and rotated sessions won't propagate. It removes the browser round-trip for authenticated reads but leaves session hygiene in your hands.

Apify tightens the orchestration layer

Apify's SDK update is housekeeping for people running Actors inside Actors. The headline is a spend ceiling on task calls: Actor.start_task and Actor.call_task now accept max_total_charge_usd, and the older max_items is deprecated in its favor — a more sensible guardrail now that much of the platform bills by usage rather than item count. The same update adds Actor.child_runs, with clients for named child runs, reattaches those named runs after migration or resurrection, and aborts them when the parent is gracefully aborted. Add Python 3.15 support and it's a release aimed squarely at teams whose Actors fan out into subtasks and need those subtasks to survive restarts and die cleanly with the parent.

What it adds up to

The through-line this week is agents getting safer, more authenticated access to data they previously couldn't reach: Firecrawl shipping datasets that arrive already vetted or already filtered, Browserbase punching through login walls without a full browser session, and Apify putting a dollar cap on the fan-out. Browser automation is increasingly the plumbing under data products rather than the product itself.