AI browser sandbox

Kernel

Kernel is open-source browser infrastructure for AI agents, launching cloud browsers fast and billing by GB-second ($0.0000166667/GB-s) with no charge for idle time or proxies.

Last verified 2026-08-09

Kernel provides fast, open-source cloud browser infrastructure plus an agent runtime, working with Playwright, Puppeteer and Browser Use. Pricing is memory-based at $0.0000166667 per GB-second with no idle or proxy charges, and plans run from Free through Hobbyist ($30/month) and Start-up ($200/month, with GPU acceleration). It has raised a $22M Series A.

Key specs

IsolationCloud browser (isolated)
Pricing modelPer GB-second (no idle/proxy charges) + plan
GPUNo
Session limitReusable sessions
PersistenceReusable sessions; session replays
MCP supportn/a
DeploymentManaged cloud + open source
Free tierFree ($5/mo credits, 5 concurrent browsers)
Pricing$0.0000166667/GB-second
PlansFree, $30, $200, enterprise
Open sourceYes

Strengths

  • Open source and framework-agnostic
  • GB-second billing with no idle or proxy charges
  • Fast launch and autoscaling
  • Agent runtime included

Trade-offs

  • Browser infrastructure, not a code-execution sandbox
  • Newer platform

Best for: Teams wanting fast, open-source cloud browsers with memory-based, idle-free pricing.

How Kernel works

Kernel is open-source browser infrastructure plus an agent runtime: it launches cloud browsers quickly and lets you connect a web agent built with Playwright, Puppeteer or Browser Use, handling the hard parts (fast spin-up, live view, logs, headless or headful mode, residential proxies, autoscaling). Its pricing model is unusual for the category, memory-based at $0.0000166667 per GB-second with no charge for idle time or proxies, which suits bursty agent workloads that would otherwise pay for waiting.

Pricing in practice & watch-outs

Plans run from Free ($5/month in credits, 5 concurrent browsers) through Hobbyist at $30/month and Start-up at $200/month (150 concurrent, GPU acceleration), all on top of the GB-second usage rate. Charging by memory-second with no idle or proxy fees can be markedly cheaper for agents that spend time thinking between actions, though it makes back-of-envelope cost comparisons against per-browser-hour rivals less direct. It is browser infrastructure rather than a code-execution sandbox, and the platform is relatively new despite strong early traction.

Head-to-head

Alternatives to Kernel

See the full Kernel alternatives comparison →

Sources: Kernel pricing ↗ · Kernel ↗

Is Kernel right for your workload?

Tell us your use case and we'll confirm the fit, flag anything to watch for, and get you real pricing from the Kernel team. No spam.