Infrastructure for agents

The runtime every agent quietly runs on.

One API call gives your agent a fresh, isolated browser — with clean IPs, vaulted credentials, and a full session recording you can replay when something goes wrong.

Trusted by teams shipping agents to production

Lindwall AIDecagraphCresta LabsOperator CoNorthwind BankHelm Health
The problem

Spinning up Chromium is the easy 20%.

Every team that ships an agent to paying customers hits the same four walls. We built the runtime for the other 80%.

01

Non-determinism

The same prompt against the same site produces different DOM paths on different runs. Customers stop trusting the agent after the third flaky failure.

02

Isolation & secrets

You cannot run a customer's banking session in the same Chromium process as another customer's. Real production needs per-session microVMs.

03

Observability

When the agent fails at step 47 of a 60-step workflow, you need video, DOM diff, and network trace of that exact session — not print statements.

04

Compliance

SOC 2, HIPAA, and SOX buyers will not let an open-source library touch their data. They need audit trails, residency, and SLAs.

What you get

Browsers that behave in production.

Every default in the runtime is the one you wish you had picked after your first outage.

Isolation

Per-session microVMs

Every session runs inside its own Firecracker microVM. No shared Chromium profiles, no cross-tenant state, no surprise cookie bleed.

Credentials

Secrets that stay secret

Store logins in the vault and reference them by handle. Values are injected at the keystroke layer — never in prompts, logs, or replays.

Replay

Full session replay

Video, DOM diff stream, HAR network log, and console output, all tied to the agent's step ID. Reproduce any failure deterministically.

Network

Clean IPs on demand

Datacenter, residential, or mobile IP pools, selectable per session. No more agents blocked on turn two by a bot filter.

Handoff

Human-in-the-loop

Webhook fires when the agent gets stuck. An operator takes over the live session in the browser, then hands control back.

Compliance

Auditable by default

SOC 2 Type II and HIPAA. Region-pinned data residency. Immutable audit logs your security team can actually point at.

One line of code

A drop-in replacement for whatever you're running today.

We ship official adapters for Browser Use, Playwright, LangChain, and OpenAI Computer Use. Point your existing stack at our runtime — your code does not change.

Read the adapter docs →
from browser_use import Agent
from browseranvil import Runtime

runtime = Runtime(api_key="ba_live_...")

agent = Agent(
    task="File my Q3 sales tax return",
    browser=runtime.session(
        region="us-west",
        credentials=["state-portal-login"],
        record=True,
    ),
)

await agent.run()
By the numbers

Built for scale, measured in production.

99.95%
Session start SLA
<900ms
Cold start, p95
12
Regions available
100%
Replays retained for audit
How it works

Four moving parts. Zero you have to maintain.

Request a session

One HTTP POST returns a Chrome DevTools WebSocket. Pick region, IP pool, and recording.

Boot a microVM

Firecracker spins a dedicated Chromium in under a second from our pre-warmed pool.

Run your agent

Your existing framework drives the browser over CDP. We stream credentials and record.

Replay & audit

Every click, keystroke, and network call is saved. Share the replay link with your team.

Use cases

Running somewhere right now.

Customer support agents

CRMs, ticketing systems, legacy account portals — all behind logins, all flaky, all now audit-friendly.

Back-office automation

Invoice reconciliation, vendor onboarding, tax filings, bank portals. The spreadsheets no one wants to maintain.

Research & data collection

Scrape respectfully at scale. Rotate IPs, pin fingerprints, and replay the exact page you extracted from.

Consumer productivity agents

Personal assistants that log into the user's own accounts — safely, with secrets never leaving the vault.

Sales & prospecting

LinkedIn, Apollo, any modern SaaS tool. Session rotation plus residential IPs keep accounts healthy.

QA & testing swarms

Agent-driven regression testing of your own app in thousands of parallel isolated browsers.

We rebuilt our agent three times before we gave up and bought a runtime. Would have shipped six months earlier on BrowserAnvil.

— Priya Natarajan, Staff Engineer, Decagraph

Works with your stack

The Switzerland of agent runtimes.

Model-agnostic by design. Use us under whatever framework you already ship.

Browser Use Playwright Puppeteer LangChain LangGraph OpenAI Computer Use Anthropic Computer Use CrewAI AutoGen Llama Index OpenTelemetry Datadog
FAQ

Answers to the questions your CTO will ask.

Isn't this just Browserbase or E2B?

We overlap on the surface: all three give you a remote browser. What is different is replay and compliance. Every session on BrowserAnvil is recorded by default, deterministically rerunnable, and backed by SOC 2 and HIPAA — features the others either sell as add-ons or do not have.

What happens if OpenAI or Anthropic offer this natively?

They might. Our hedge is being rigorously model-agnostic and infrastructure-first. The value of compliance, replay, and per-session microVMs does not go away because the model gets better — if anything, it becomes more important.

Can I self-host?

Enterprise customers can deploy a dedicated tenant in their own VPC. Pure self-hosting for the control plane is on the roadmap for 2026.

How do you handle bot detection and CAPTCHAs?

Residential IP pools, stable fingerprints, and — when needed — human handoff via webhook. We do not solve CAPTCHAs for you, but we make it trivial to pause the agent and hand the session to an operator.

What is the free tier?

60 session-minutes every month, forever, with every production feature enabled. We want you to ship your first demo without pulling out a card.

Ship an agent your compliance team will sign off on.

Free tier includes 60 session-minutes a month. No credit card.

Start free Book a demo