Kernel is browser infrastructure as a service: it lets an AI agent spin up a cloud browser without a team building that layer themselves.
What it is
Kernel's REST API exposes the full lifecycle of a browser session: pools, profiles, proxies, replays, extensions, and computer-use controls for mouse, keyboard, clipboard, and screenshots, plus an in-browser filesystem and process execution. It launches cloud browsers in under 30ms, with GPU acceleration available, and can reuse session state like cookies and authentication so an agent's context persists across a workflow instead of starting cold every run. The company raised $22M, reported in October 2025, to build this out, going head-to-head with Browserbase on launch speed and parallel session handling.
How it gets used to scrape
Kernel's pitch is speed and concurrency: agents that need many browser sessions running in parallel, not one long-lived session. That fits high-throughput scraping and automation jobs where the bottleneck is spinning up enough browsers fast enough, plus workflows that need persisted auth and cookies across repeated runs rather than logging in fresh each time.
Why it's hard to catch with old checks
Kernel advertises built-in anti-detection as part of its browser infrastructure, on top of the raw speed. That combination means the fingerprint checks a scraper would otherwise have to patch in manually are handled by the platform instead. It also means the traffic shares infrastructure across many different customers' agents, so a check that only looks at one session's fingerprint in isolation misses the pattern that shows up across Kernel's fleet.
Centinel detects it
In our September 2026 benchmark, Centinel identified every session we ran through Kernel. Fast launch times and platform-level anti-detection change how quickly a session spins up, not how it behaves once it's running, and behavior across the request, browser, and connection layers is what Centinel scores.
Key takeaways
- Kernel is browser infrastructure as a service, built around sub-30ms launch times, GPU acceleration, and persisted session state across runs.
- It's built for high-concurrency agent workloads, not single long-lived sessions.
- Built-in anti-detection changes the fingerprint a session presents, not the behavioral pattern Centinel scores across the session.