Loading page...Загружаем страницу...正在加载页面...
AtlasRepo

Browser Use vs Skyvern Self‑Hosted: Setup, Trade‑offs, and Completion Verification

Description and use

Evaluation of Browser Use and Skyvern revealed that agent‑reported completion can be misleading without independent artifact verification. Browser Use is a lightweight component; Skyvern is a full‑stack platform.

Problem

Determine whether Browser Use or Skyvern can reliably complete a recurring login‑to‑download workflow without creating a permanent human repair queue.

Outcome

Both tools can execute the workflow, but verification requires independent artifact checks; Browser Use offers a lightweight embeddable component, while Skyvern provides a broader platform at higher engineering cost.

Workflow

  • Install Browser Use via uv and pip, ensuring Chromium is provisioned explicitly.
  • Write a minimal Agent script that logs in, fills a three‑stage form, submits, downloads a receipt, and returns its filename and ID.
  • Install Skyvern via git clone, copy .env.example, and run docker compose up, ensuring the server‑capable package is used.
  • Define a Pydantic 2.10.6 model for DownloadReceipt and implement verify_artifact to check file size and SHA‑256 digest against expected values.
  • Run both systems against the same demo application and compare verified completion rates, wall‑clock time, model requests, retries, and human interventions.
  • Document that structured output alone can falsely mark a workflow complete.

Evidence

We did not bring Browser Use and Skyvern into our lab to see whether an agent could click a button in a polished demo. We wanted to answer a more expensive question: can either system complete a recurring login-to-download workflow without creating a permanent human repair queue?

Source

Agent‑reported done is not verified completion

Source

A strict schema receipt still passed for a missing and a corrupted file, so a workflow should count as complete only when page state, business identifier, and artifact size and digest are checked independently of the agent's own success claim.

Source

For Browser Use, we use Python 3.11 or newer and take uv add browser‑use as the documented setup path.

Source

For Skyvern, we used the full‑stack deployment boundary rather than mistaking the lightweight Python package for the server.

Source

  • Record date: 2026-10-04T01:40:51.531Z