BrowserTrace

Which tool should you use to debug an AI browser-agent failure?

BrowserTrace is intentionally narrow: it is a local flight recorder for the moment an AI browser agent fails. It complements Playwright Trace Viewer, LLM observability tools, and hosted browser platforms rather than replacing them.

View repo Open exported trace Playwright + LLM guide

Short answer

Debugging job Best fit Where BrowserTrace fits
You need deep Playwright test trace inspection: actions, DOM snapshots, console logs, network events, and timing. Playwright Trace Viewer Use BrowserTrace when the same Playwright run also includes an LLM prompt, model output, selected action, and browser-agent failure story.
You need LLM application tracing, monitoring, evaluations, prompt iteration, cost, latency, and production dashboards. LangSmith or Langfuse Use BrowserTrace when the missing context is what the browser looked like, where it was, what action the agent tried, and which step first failed.
You need hosted browser infrastructure, session replay, live debugging, and managed browser sessions. Browserbase Use BrowserTrace when you want a local, runtime-agnostic trace that works with your own Playwright, Browser Use, Stagehand, Skyvern, or custom code.
You need a small OSS artifact to attach to a bug report, issue, pull request, or teammate review. BrowserTrace Export a standalone HTML trace with screenshots, URL, action, model input/output, status, errors, and optional redaction.

BrowserTrace vs Playwright Trace Viewer

Playwright Trace Viewer is the right tool for inspecting Playwright automation itself. BrowserTrace is for the layer above that: an LLM or browser agent chose an action, the browser state changed, and you need a compact timeline that explains the failed agent decision.

Use both together when needed: Playwright Trace Viewer for low-level browser mechanics, BrowserTrace for the agent-facing failure report.

BrowserTrace vs LangSmith or Langfuse

LangSmith and Langfuse are better fits for broad LLM observability: tracing calls, monitoring production behavior, evaluating outputs, comparing runs, and analyzing cost or latency. BrowserTrace is smaller and more specific. It stores the visual browser context beside the model decision so a failed browser-agent step is easier to inspect locally.

A common setup is to keep your LLM observability tool for production traces and add BrowserTrace around local browser-agent debugging or shareable bug reports.

BrowserTrace vs Browserbase recordings

Browserbase is a hosted browser platform with session observability and live debugging. BrowserTrace is not a hosted browser runtime. It runs with the browser stack you already use and writes local traces that can be exported as HTML.

Use Browserbase when you want managed browser sessions. Use BrowserTrace when you want a lightweight OSS trace around a local or self-managed browser-agent run.

What BrowserTrace records

If your browser agent failed and a normal log does not show what it saw, start with the exported trace demo or run browsertrace demo locally.