What is Kernel Browsers?
Kernel Browsers is browser infrastructure for web agents, customer automations, and scraping jobs. Kernel's own documentation calls its user-directed browser identity Kernel Agent and says it visits pages in response to an end user's request. It is not an automatic crawler with one shared indexing mission.
These sessions use a browser, so they may navigate, interact with controls, or operate inside an authenticated workflow when the customer has legitimate access. A Kernel visit can be part of many different tasks. The platform identity does not reveal the customer, the instruction, or what happens to retrieved material afterward.
Kernel documents a separate identity for Kernel Search. That crawler uses the KernelSearchBot token to build search indexes and retrieval databases, and it follows robots.txt for that token. This page covers the broader Kernel browser agent, whose record does not contain a stable user-agent token. The two identities should not be combined when writing access rules.
Nothing in the Kernel Browsers record says that these sessions collect a shared foundation-model training set or maintain a public AI search index. A customer's own agent may use page content during a task, but allowing Kernel does not by itself improve general AI search visibility. Blocking it can instead prevent a requested automation from completing.
Kernel identifies its agent with Web Bot Auth. Signed requests can be checked against https://www.kernel.sh/.well-known/http-message-signatures-directory, which Cloudflare also lists for the Kernel agent. This is stronger evidence than a browser user agent because the HTTP message signature can be verified cryptographically.
Robots behavior depends on which Kernel identity is present. Kernel promises robots compliance for KernelSearchBot, while Cloudflare's bot directory marks the user-directed Kernel agent as not following robots.txt. Since the browser agent has no dedicated token, control it by verified signature, authentication, and edge policy rather than inventing a robots.txt name.
