The teams shipping the best agent products — Cognition, Lindy, Cresta, Decagon — all built this runtime in-house because it did not exist as a product. Now it does.
Every agent team we talk to has lived the same story. Here is the short version.
Works for the demo. Breaks on the third customer because they share an IP.
Flakiness drops. Support tickets about "the agent typed my password into the wrong form" start arriving.
Two engineers for six weeks. It works. It is not SOC 2-ready.
print() statements, then screenshots, then an in-house replay tool that half-works.
A bug leaks one customer's cookies into another's session. Emergency pivot to microVMs.
The first enterprise deal stalls in security review. You hire a compliance consultant.
Or you can skip straight to quarter seven.
Swap your browser provider for BrowserAnvil. Your agent code does not change. What changes is the failure rate, the debuggability, and the answer you give when a customer's security team asks how you handle isolation.
- browser = chromium.launch()
+ browser = browseranvil.session(
+ credentials=["customer-okta"],
+ ip_pool="residential",
+ record=True,
+ )
# rest of your agent unchangedPre-warming, eviction, crash recovery, version upgrades. All ours.
Residential, mobile, datacenter pools with health checking and reputation management.
Encrypted storage, rotation, TOTP, WebAuthn proxy.
Video, DOM diff, HAR, and the viewer to scrub it all.
SOC 2 Type II, HIPAA, ISO 27001 — inherited, not re-earned.
Our engineers get paged at 3 a.m., not yours.
Pick Python or TypeScript. Import the adapter for your framework.
Feature-flag a small fraction of agent runs through BrowserAnvil. Compare success rates.
Migrate production secrets. Enable auto-redaction in logs and replays.
Turn down the old Chromium fleet. Keep it on standby for a week, then delete.
With SOC 2, replay, and compliance in the first slide.
We moved off our self-hosted Chromium fleet on a Tuesday. By Friday our on-call pages for browser flakiness were down 80%.
Three engineers, two demos, and an infra rabbit hole that is eating the roadmap. We cut it in half.
Paying customers, but incidents, and a security-review queue that is stalling deals. We unblock both.
A central team serving other teams at a larger company. We give you the building blocks, not another framework to own.
Real users, real accounts, real stakes. Per-user isolation matters; replay matters more.
Almost certainly. We ship official adapters for Browser Use, Playwright, LangChain, and Computer Use. Anything CDP-compatible works unchanged.
Most teams have a real workload running on us within an afternoon. A proper production cutover tends to take a week.
Keep it if you love it — ours supplements rather than replaces. Most teams move fully over within a month.
Your first sixty minutes are free, forever.