BrowserTrace
VerifiedLocal-first debugger for replaying browser agent runs and failures.
What is BrowserTrace?
BrowserTrace is an open-source local replay tool that records every step taken by browser-based AI agents. It stores screenshots, navigation details, actions, and model responses in a local database so developers can review exactly what happened during a run.
Users open a local interface to jump to any recorded step, compare a failed run against a successful one, or export sanitized traces for sharing. The system works entirely offline and focuses on helping identify the first point where an agent diverged from expected behavior.
It is intended for developers and researchers who build or debug browser agents and need concrete visibility into visual state, model decisions, and execution errors without relying on external logging services.
Capabilities
What you can build with BrowserTrace
Debugging upload failures
Review the exact navigation and preview steps when an agent attempts to upload a local file and ends up on the wrong page.
Comparing runs
Run a side-by-side comparison between a failed task and a known-good execution to surface the first differing action or error.
Multi-framework tracing
Record sessions from Browser Use, Stagehand, Skyvern, or Playwright plus LLM scripts in the same local timeline format.
Install BrowserTrace
uvx --from "browsertrace[ui]" browsertraceuvx --from "browsertrace[ui]" browsertrace doctor
uvx --from "browsertrace[ui]" browsertrace demo
uvx --from "browsertrace[ui]" browsertrace list
uvx --from "browsertrace[ui]" browsertrace- 1Install uv if it is not already present on your system.
- 2Run the doctor command with uvx to verify the environment.
- 3Execute the demo command to generate a sample trace.
- 4Use the list command to view recorded runs in the terminal.
- 5Launch the local UI to inspect screenshots and model outputs.
BrowserTrace: pros & cons
Pros
- +Fully local with no cloud dependency or sign-up required
- +Provides visual replay of screenshots alongside model I/O
- +Supports comparison between failed and successful runs
- +Works with multiple browser agent frameworks beyond the primary one
Cons
- –Requires Python 3.11 or newer and uv or pip for installation
- –UI features need the optional [ui] extra to be installed
- –Primarily designed around browser automation rather than general agents
Frequently asked questions
No, all recording and storage happens locally on the user's machine.
User reviews
Verified reviews from the community shape this listing's rating.
Loading reviews…
Similar agents
Other web & browser options worth comparing.
