Where does the browser automation code run?
An external worker, a hosted browser Function, a browser-local app action and a hosted agent are different execution models. The choice determines who deploys the code, starts the browser, records the result and stops the run.
A scheduled job should return a verified result and stop its owned browser even when an action fails. The scheduler's successful HTTP trigger is not proof that the browser completed the business operation.
Compare the execution model
Kernel: Deploy a TypeScript or Python App and expose Actions that callers invoke. Kernel co-locates code with the browser environments the app creates. An Action invocation is the execution unit rather than a bare CDP connection.
Anchor Browser: Automation Tasks produce versioned workflows from a prompt or a recorded demonstration. Test and edit the workflow, then invoke it through the task API. Compare this authoring model with writing a handler directly.
Compare deployment ownership
Kernel: Kernel Apps move the automation process next to the browser. Alternatively, an external process can control a browser through the browser API. Choose based on execution locality and deployment ownership.
Anchor Browser: Use a browser session from existing code or deploy a reusable Automation Task. The task path includes workflow editing and versioning; direct CDP leaves execution ownership with your application.
Compare the evidence from a run
Kernel: Live browser URLs support an embedded view and a read-only option. Replays and browser telemetry are separate observability surfaces; app invocation logs add execution context.
Anchor Browser: Anchor documents MP4 session recordings and a live browser used during demonstrations. Keep the session artifact alongside the task's version, inputs and completion result.
How to evaluate this choice
Keep an external runner when you already operate one and need portability. A native execution layer can remove deployment work; browser-local code can reduce repeated network calls. Compare actual workload behavior before assuming either always wins.
- Confirm the supported language, entry point and dependency packaging.
- Trace deployment, version selection, invocation and cancellation.
- Keep the handler runtime limit separate from browser session duration.
- Decide whether scheduling is native or supplied by another service.
- Associate results and logs with both the code version and browser session.
Requirements side by side
| Requirement | Kernel | Anchor Browser |
|---|---|---|
| Code execution | TypeScript / Python Apps | Automation Tasks |
| Scheduled execution | App invocation path | Task execution path |
| Recordings / replay | Documented | Documented |
Sources: kernel.sh [1] ↗kernel.sh [2] ↗kernel.sh [3] ↗kernel.sh [4] ↗kernel.sh [5] ↗kernel.sh [6] ↗docs.anchorbrowser.io [7] ↗docs.anchorbrowser.io [8] ↗docs.anchorbrowser.io [9] ↗docs.anchorbrowser.io [10] ↗docs.anchorbrowser.io [11] ↗docs.anchorbrowser.io [12] ↗
Which approach fits?
Choose Kernel when your code makes many browser calls and you want to deploy it beside the browser, or you need Kernel's managed sign-in flow.
Kernel tradeoff: Browser-local code changes the deployment model. Check app action behavior, account limits and persistent-state ownership before moving an external worker into it.
Choose Anchor Browser when you want managed sign-in and an automation task you can demonstrate, edit and invoke from your application.
Anchor Browser tradeoff: A demonstrated automation task and an arbitrary Python handler are different authoring models. Evaluate how much control you need over the code, dependencies and exact workflow.
A useful acceptance check
Write a pass/fail result for the real task before running the evaluation. Record the account, configuration and code version next to that result. Repeat failures need an explanation, not an automatically different browser configuration.
When Notte belongs on this shortlist
Kernel's browser-local Apps make execution placement part of the decision. Compare the repeated action path with the proposed Notte controller or Function deployment before concluding that a lifecycle or hourly-rate difference decides the workflow.
Anchor's demonstration or prompt-authored Automation Tasks and Notte's Python Functions suit different authoring preferences. Both document website-authentication tooling, so a credible recommendation must address the actual login and workflow-editing path.
Notte: Functions deploy Python code with a run() entry point. Invoke them over HTTP or with an SDK, inspect run history, and roll back code by redeploying an earlier version as the latest version.
Consider this third option when its documented mechanism addresses a requirement left open by the two providers above. A working configuration on either original provider can still be the better choice; compare the task and ownership cost before switching.
Sources: kernel.sh [1] ↗docs.anchorbrowser.io [2] ↗docs.notte.cc [3] ↗