Changelog

Browser requests were silently going through the non-JS engine for 41 days. If you saw Browser fail on Cloudflare challenges or JS-heavy sites during that window, this was why. We fixed the routing. Browser now executes JavaScript on every request again.

The product chip filter (Single, Proxy, Browser) now works on Metrics and Activity too, not just Overview. Pick a product anywhere, and your choice carries across all three pages. One small caveat: the Concurrency scope on Metrics is disabled when a product filter is active, since concurrency isn't broken out by product yet.

Added an explicit acceptable-use clause covering third-party websites, plus a clearer jurisdiction section: Sofia courts with the EU consumer carve-out preserved. Worth a quick re-read if you're working with us at scale.

The dashboard preview on the homepage was cramped on small screens. We fixed the type and spacing, and added tooltips that actually work on touch (hover-only hints don't help a phone).

We updated robots.txt with a clear stance for 2026. Search engines and social previewers (Facebook, LinkedIn, Twitter, Slack, Discord and friends) are explicitly welcome. AI training crawlers are not.

Sharing a FourA link on Twitter, Slack, LinkedIn, or Discord used to fall back to one generic card. Every public page now has its own preview image with the right title and chips. Small detail, but it shows up every time someone shares us.

We just put up a /jobs page with two open positions. If you like building real systems instead of stitching together SaaS, take a look. We work in-house on every layer, ship constantly, and treat OSS as a first-class neighbour.

We found two ways a dead proxy could get pinned in the picker and never get evicted, which dragged response times under sustained load. Both are fixed. In our load tests, p95 dropped from over 5 seconds back to under 500ms and the fail rate went from 8% to zero.

Smart picker, round two. The working pool used to grow as we discovered new proxies, and slower ones snuck back into rotation. Now it drops them on the fly, so response times hold steady whether you've been running ten minutes or ten hours.

Proxy selection used to be random. Now Proxy Finder remembers which proxies performed well for each destination, and picks those first. Early requests still sample a few to build the picture. After that, you should see steadier response times and fewer slow tails on repeat targets.

We tightened up our rollouts. When we ship a new version of Single, Proxy Finder, or Browser, the load balancer waits for each new instance to be actually ready before routing traffic to it. The brief blips you might've seen during our release windows are gone.

Country and continent fields on proxy lookups are more accurate now. We promoted the heavier-weight geo data source to authoritative and dropped the lighter one to fallback. Border-region IPs that used to flip between countries stay put. Filter by country in Proxy Finder, and your results should be steadier.

If your sign-in used to land on a raw 400 page after a stale tab or a slow callback, that's fixed now. We detect the session mismatch, wipe the bad cookie, and send you back to the login screen. Cleanly, every time.

New proxies stay inactive in Proxy Finder until validation confirms they actually work. Before this fix, fresh imports could be picked for live requests before anyone confirmed they were responding, which dragged down first-attempt success rates. That path is closed now.

We pruned two dead source feeds and added 25 new ones. Input validation got tighter too, so non-IPv4 hosts and unknown protocols never enter the pool. The result is a larger, cleaner pool feeding Proxy Finder.