Browserbase on X

Plain praise is the largest theme, and it runs brief and unreserved: "is the goat", "Top Notch", quick recommendations, and one admiring description of splitting browser tasks into executors and teachers. Behind it, people describe work actually done with it: cleaning up 13 years of tweets, paying 0.01 USDC per fetch through x402 on Base, search inside agent routing, proxy rules that are easier to manage and update, and receipt processing reported at 5x with more than 30,000 manual hours saved and browser input cut from about 122,000 tokens to 4,000.

The capability people single out is login handling: approving a sign-in in 1Password so the agent never sees the credential, or handing the live browser to the human for a one-time code — "the most sensible agent feature i've seen". 5 people raise it. Two credit a forked Chromium and signed requests for how sites read the agent, and two say they are only planning to try it.

Cost is one of the two things people raise: it is a paid product, and a per-session charge sits on top of everything else — one person counts a virtual card fee, interchange and a retry charge stacked on a single agent purchase. Another contrasts it with self-hosting Steel or Langfuse to cut cost.

Session fragility is level with cost. One detailed report says browsers wipe cookies about 0.3 seconds after a credentialed fetch gets a cross-site 302, breaking a login with "Page expired" on the same Chrome that works locally; one person notes a third party can see your sessions; another says that for personal agents, what sits behind a sign-in is the real problem.

Two people say cloud browsers get blocked where a real browser wouldn't — captchas, and a test where it "stopped working just before he recorded". Two describe runs that look successful when they aren't: product pages returning $0 because the agent never scrolled, or a click reported done with the state unchanged.

Steel and Exa are the two names with more than one person behind them, and they stand for different worries. Steel is about who runs the box: self-hosting Steel cuts cost, while Browserbase "sells managed polish", and one person counting the cost of agent checkout puts a headless browser session — "browserbase or steel or whatever" — as another cost on top of the card.

Exa is the question of overlap: whether Browserbase duplicates what Exa and web search already do, or runs alongside them, as in a proposed search waterfall "across Exa, Parallel, and Browserbase".

The single-person comparisons read as fit rather than fault. Firecrawl is described as scraping and indexing APIs for feeding agents web data, Browserbase as cloud sessions for long or concurrent web tasks. Browser Use's Profile Sync and Browserbase Contexts are weighed for keeping login state across remote sessions. Camoufox is weighed as an open-source browser on the agent's own machine against the hosted one. And in the most-reached post in the section, one person tells others to use Browserbase "instead of self-hosting Chrome on VPS (memory hog)".

Getting through a site is the biggest question, and the specifics are concrete: how to pass agent blocks on OpenTable, how to handle a two-factor code that arrives halfway through a booking, whether the agent scrolls by default or waits on a selector when lazy-loaded prices come back at $0, and which failure modes a skills layer actually cut — with one person saying the brittle part was "the acceptance check after the click".

Overlap is next: whether it does "the same thing that Browserbase and Exa already do", whether a search and research waterfall across Exa, Parallel and Browserbase makes sense, and a plain "doesnt browserbase do the same?"

Price questions ask why a subscription is needed at all and what one search costs after the first thousand, which one person says decides whether search stays inside the browser loop. Proxy questions ask whether rules scope by domain or by tool name and whether a tier includes residential proxies. Singly, someone asks if the search returns images, and a subscriber with usage left asks what else people use it for.

The clearer of the two asks is for an "individual" plan at around $10 a month for 10 to 20 hours of browser use — wanted "purely for personal agent integration", by someone who says they don't want it free but aren't developing an app, and expects more people to want the same as their agents build things for them.

The other is a short reply asking for Browserbase support for Duo, a few words with no reason given.

These are the only two requests in 2 posts from 2 people, so they are single asks rather than anything a group is pushing for. The personal-use plan is spelled out with a price and a usage range; the Duo request names a tool and nothing more.