BrowserTrace Examples

May 13, 2026 · View on GitHub

Use these examples to create local traces before wiring BrowserTrace into a real agent. The no-service examples do not require API keys, hosted browser accounts, or network access.

Fastest Path

The quickest no-install path is uvx from PyPI:

uvx --from "browsertrace[ui]" browsertrace doctor
uvx --from "browsertrace[ui]" browsertrace demo
uvx --from "browsertrace[ui]" browsertrace list
uvx --from "browsertrace[ui]" browsertrace

Check a healthy local install

browsertrace doctor prints the local install and trace-store status before you open the UI:

BrowserTrace doctor
Python: 3.11.x
Home: /tmp/browsertrace-demo
Database: /tmp/browsertrace-demo/db.sqlite
Runs: 0
Steps: 0
Next: browsertrace demo
UI dependencies: available

Windows PowerShell:

$env:BROWSERTRACE_HOME = "$env:TEMP\browsertrace-demo"
uvx --from "browsertrace[ui]" browsertrace doctor
uvx --from "browsertrace[ui]" browsertrace demo
uvx --from "browsertrace[ui]" browsertrace

Persistent install from PyPI:

pip install "browsertrace[ui]"
browsertrace doctor
browsertrace demo
browsertrace

Open http://127.0.0.1:3000 and inspect demo: Browser Use local HTML upload navigation failure. From a source checkout, python examples/no_api_failure_demo.py creates the same trace.

For a downloadable public-safe export that omits prompt/model I/O, screenshots, and URLs, inspect the release asset:

https://github.com/aaronlab/browsertrace/releases/download/v0.1.17/browsertrace-demo-public.html

Discover CLI options

Use the built-in help when you want to inspect available commands or export flags from the terminal:

browsertrace --help
browsertrace export --help
browsertrace export <run_id> --public -o public.html

A healthy browsertrace --help output should list the main commands, including serve, doctor, demo, list, show, compare, and export. browsertrace export --help should mention public-safe export options such as --public and an output path like -o public.html or --out public.html.

BrowserTrace command cheat sheet

Command	Use when
`browsertrace doctor`	Check the local install and trace-store status
`browsertrace demo`	Create a deterministic failed run
`browsertrace list`	Find recent run IDs
`browsertrace show <run_id>`	Inspect a run timeline in the terminal
`browsertrace compare <failed_run_id> <success_run_id>`	Find the first step divergence between a failed run and a known-good run
`browsertrace export <run_id> --public -o public.html`	Create a public-safe HTML export

Example compare output:

$ browsertrace compare failed-local-html-upload good-local-html-upload
First divergent step: 3

failed:
  action: navigate
  url: file:///tmp/browsertrace-report.html
  status: failed
  error: upload preview did not appear

good:
  action: upload_file
  url: https://example.test/upload
  status: ok

Use browsertrace compare <failed_run_id> <success_run_id> --json when a script, CI check, or AI/coding-agent troubleshooting flow needs structured comparison output.

Example Matrix

Example	Use when	Extra services	Command
`no_api_failure_demo.py`	You want the fastest failing trace	None	`python examples/no_api_failure_demo.py`
`browser_use_callback_demo.py`	You want to see Browser Use-shaped step callbacks recorded	None	`python examples/browser_use_callback_demo.py`
`browser_use_run_hooks_demo.py`	You want to see Browser Use `agent.run(...)` lifecycle hooks recorded	None	`python examples/browser_use_run_hooks_demo.py`
`stagehand_wrapper_example.py`	You want to see Stagehand-style `act` and `extract` calls recorded	None	`python examples/stagehand_wrapper_example.py`
`skyvern_wrapper_example.py`	You want to see Skyvern-style task calls recorded	None	`python examples/skyvern_wrapper_example.py`
`playwright_llm_loop_example.py`	You want Playwright + LLM-shaped prompt, DOM, selector, retry, and failure fields without a browser	None	`python examples/playwright_llm_loop_example.py`
`computer_use_loop_example.py`	You want a generic observe-decide-act computer-use trace	None	`python examples/computer_use_loop_example.py`
`basic_example.py`	You want the smallest manual SDK example	None	`python examples/basic_example.py`
`failure_example.py`	You want a deterministic failed run with richer steps	None	`python examples/failure_example.py`
`playwright_example.py`	You have Playwright installed and want browser screenshots	Local Chromium	`python examples/playwright_example.py`
`multipage_failure.py`	You want a multi-page Playwright failure demo	Local Chromium	`python examples/multipage_failure.py`
`browseruse_example.py`	You already use Browser Use and want to attach tracing	Browser Use + LLM	`python examples/browseruse_example.py`

Stack-specific debugging guides

Use the guide closest to the example or agent stack you are trying:

Browser Use guide: https://aaronlab.github.io/browsertrace/browser-use-debugging.html
Stagehand guide: https://aaronlab.github.io/browsertrace/stagehand-debugging.html
Skyvern guide: https://aaronlab.github.io/browsertrace/skyvern-debugging.html
Playwright + LLM guide: https://aaronlab.github.io/browsertrace/playwright-llm-debugging.html
Computer-use guide: https://aaronlab.github.io/browsertrace/computer-use-agent-debugging.html
Failure patterns overview: https://aaronlab.github.io/browsertrace/browser-agent-failure-patterns.html

For help deciding when to use BrowserTrace alongside other debugging tools, see the comparison guide: https://aaronlab.github.io/browsertrace/compare-browser-agent-debugging.html

Browser Use users who pass run hooks directly to agent.run(on_step_start=..., on_step_end=...) should use create_run_hooks; see the Browser Use debugging guide for the current run-hook path. To try that path without installing Browser Use, run python examples/browser_use_run_hooks_demo.py from a source checkout.

If you are adapting the no-service wrapper examples, see the Stagehand debugging guide or the Skyvern debugging guide for the current integration paths.

For the no-service Playwright + LLM and custom computer-use loops, see the Playwright + LLM debugging guide or the computer-use debugging guide for the current debugging flow and field coverage.

Quick local verification for a no-service wrapper example

For a compact first-run check, the Stagehand wrapper example should create one successful local trace that you can verify from the terminal before opening the UI:

BROWSERTRACE_HOME=/tmp/browsertrace-demo python examples/stagehand_wrapper_example.py
BROWSERTRACE_HOME=/tmp/browsertrace-demo browsertrace list --limit 5
BROWSERTRACE_HOME=/tmp/browsertrace-demo browsertrace show <run_id>

Expect browsertrace list --limit 5 to show a recent demo: stagehand checkout flow run, and browsertrace show <run_id> to list two successful steps: act: click the checkout button and extract: extract the order total.

For a no-service Browser Use-shaped callback demo, run:

BROWSERTRACE_HOME=/tmp/browsertrace-demo python examples/browser_use_callback_demo.py
BROWSERTRACE_HOME=/tmp/browsertrace-demo browsertrace list --limit 5
BROWSERTRACE_HOME=/tmp/browsertrace-demo browsertrace show <run_id>

Expect browsertrace list --limit 5 to show a recent demo: browser-use callback flow run, and browsertrace show <run_id> to list two successful steps: search_google(query=BrowserTrace) and click(selector=#result-1).

For a no-service Browser Use-shaped run-hook demo, run:

BROWSERTRACE_HOME=/tmp/browsertrace-demo python examples/browser_use_run_hooks_demo.py
BROWSERTRACE_HOME=/tmp/browsertrace-demo browsertrace list --limit 5
BROWSERTRACE_HOME=/tmp/browsertrace-demo browsertrace show <run_id>

Expect browsertrace list --limit 5 to show a recent demo: browser-use run hooks flow run, and browsertrace show <run_id> to list two successful steps: search_google(query=BrowserTrace) and click(selector=#result-1).

For Playwright examples, install the browser runtime first:

pip install playwright
playwright install chromium

Contributing Example Fixes

For a docs or example fix, start with the First PR Recipe; it keeps the first contribution small and reviewable. Keep example changes narrow and include the command you used to verify the changed example or docs check.

Playwright sync API snapshot

Use snapshot_sync when your code uses sync Playwright instead of playwright.async_api:

from browsertrace import Tracer
from playwright.sync_api import sync_playwright

tracer = Tracer()

with sync_playwright() as p:
    browser = p.chromium.launch(headless=True)
    page = browser.new_page()

    with tracer.run("sync Playwright example") as run:
        page.goto("https://example.com")
        run.snapshot_sync(page, action="opened example.com")

    browser.close()

Troubleshooting

First-run troubleshooting checklist

Run these commands in order when a first run does not look right:

browsertrace doctor
browsertrace demo
browsertrace list
browsertrace show <run_id>
browsertrace export <run_id> --public -o public.html

JSON CLI checks for automation

For scripts, CI, or AI/coding-agent troubleshooting, use JSON output to check the local install, find failed runs, and inspect one run:

browsertrace doctor --json
browsertrace list --status failed --json
browsertrace show <run_id> --json

Expected output:

browsertrace doctor --json prints JSON describing your local install, configuration, and trace-store status.
browsertrace list --status failed --json prints recent failed runs as JSON, including run IDs and status values you can feed into automation.
browsertrace show <run_id> --json prints one run as JSON, including run details and step timeline data for that run.

For compact AI/coding-agent troubleshooting context, use docs/llms.txt; it includes JSON CLI checks, project links, and prompts.

Environment variable quick reference

Use these variables when you want to change local BrowserTrace behavior without editing code:

Variable	What it changes
`BROWSERTRACE_HOME`	changes the trace store directory
`BROWSERTRACE_PORT`	changes the local UI port

Example values: use BROWSERTRACE_HOME=/tmp/browsertrace-demo when you want an isolated local demo store, and use BROWSERTRACE_PORT=4000 when port 3000 is already in use.

Port Already in Use

If http://127.0.0.1:3000 is already in use, you may see an error like:

Error: listen EADDRINUSE 127.0.0.1:3000

To resolve, either:

Find the process using port 3000, then stop only that process:

lsof -nP -iTCP:3000 -sTCP:LISTEN
kill <PID>

Or run BrowserTrace on another port:

BROWSERTRACE_PORT=4000 browsertrace

Windows PowerShell:

$env:BROWSERTRACE_PORT = "4000"
browsertrace

Finding Your Demo Run

To list all local BrowserTrace demo runs:

browsertrace list

This shows run IDs, timestamps, and status for each stored trace.

Show only recent runs

Use --limit when you only need the most recent runs before choosing one to inspect or export:

browsertrace demo
browsertrace list --limit 5
browsertrace show <run_id>
browsertrace export <run_id> --public -o public.html

Inspect a failed run in the terminal

Use show when you want to confirm the failed step before opening the UI or exporting HTML:

browsertrace demo
browsertrace list
browsertrace show <run_id>
browsertrace export <run_id> --public -o public.html

browsertrace show <run_id> prints the run status, error, and step timeline in the terminal, including the failed step.

To export a trace without sensitive data (prompt/model I/O, screenshots, URLs):

browsertrace export <run_id> --public -o public.html

This creates a share-safe HTML file suitable for sharing with others.

Attach a public-safe export to an issue

When asking for help, attach public.html to a GitHub issue or PR comment and include the run status or error text from browsertrace show <run_id>. The public export omits prompt/model I/O, screenshots, and URLs.

Where Traces Are Stored

By default BrowserTrace stores SQLite data and screenshots in ~/.browsertrace/. To keep example output isolated, set BROWSERTRACE_HOME:

BROWSERTRACE_HOME=/tmp/browsertrace-demo python examples/stagehand_wrapper_example.py
BROWSERTRACE_HOME=/tmp/browsertrace-demo browsertrace

Windows PowerShell:

$env:BROWSERTRACE_HOME = "$env:TEMP\browsertrace-demo"
python examples/no_api_failure_demo.py
browsertrace

Testing with isolated trace storage

Use a temporary trace store when adding BrowserTrace to pytest suites so test runs do not write to ~/.browsertrace/. This recipe uses no browser, network, or API key:

def test_browsertrace_trace_uses_temp_store(tmp_path, monkeypatch):
    monkeypatch.setenv("BROWSERTRACE_HOME", str(tmp_path))

    from browsertrace import Tracer

    tracer = Tracer()
    with tracer.run("pytest isolated trace") as run:
        run.step(
            action="record deterministic step",
            model_input={"task": "exercise trace storage"},
            model_output={"status": "ok"},
        )

    assert (tmp_path / "db.sqlite").exists()

If your test module imports BrowserTrace before setting BROWSERTRACE_HOME, pass the temp path explicitly instead:

tracer = Tracer(home=tmp_path)

When this recipe works, the isolated store contains db.sqlite under the temporary path instead of ~/.browsertrace/. The recorded run is named pytest isolated trace, and the example does not require screenshots, a browser, network access, or API keys.

Public Export Flow

After creating a demo run, export it for safe sharing:

browsertrace demo
browsertrace list
browsertrace export <run_id> --public -o public.html

After export succeeds, public.html should appear in the current directory. Open the file locally to confirm it shows the run timeline, status, and error details while omitting prompts/model I/O, screenshots, and URLs. If the export reports an ambiguous run ID prefix, copy more characters from the full Run ID: printed by browsertrace demo, then rerun the export command.

Windows PowerShell:

browsertrace demo
browsertrace list
browsertrace export <run_id> --public -o public.html

Run ID Prefixes For Export

browsertrace list prints a short run ID prefix for each local run. browsertrace export accepts the full run ID or any unique prefix:

browsertrace list
browsertrace export <run_id> --public -o public.html

If export reports more than one matching run, copy more characters from the full Run ID: printed by browsertrace demo, then rerun the export command.

Note: --public omits prompts/model I/O, screenshots, and URLs. BrowserTrace does not upload the file anywhere; attach the generated public.html file manually to share it.

GitHub Actions artifact for public-safe exports

Use an Actions artifact when a CI job should keep a public-safe trace for review. BrowserTrace does not upload traces by itself; the upload step below is the standard GitHub artifact action.

- name: Install BrowserTrace
  run: |
    python -m pip install \
      "browsertrace[ui]"

- name: Create public-safe BrowserTrace export
  shell: bash
  run: |
    export BROWSERTRACE_HOME="$RUNNER_TEMP/browsertrace"
    RUN_ID=$(browsertrace demo | awk -F': ' '/Run ID:/ {print \$2}')
    browsertrace export "$RUN_ID" --public -o public.html

- name: Upload public-safe BrowserTrace export
  uses: actions/upload-artifact@v4
  with:
    name: browsertrace-public-export
    path: public.html

GitLab CI artifact for public-safe exports

Use GitLab CI artifacts when a pipeline should keep a public-safe trace for review. BrowserTrace does not upload traces by itself; GitLab stores the public.html file because it is listed under artifacts.

browsertrace-public-export:
  image: python:3.11
  script:
    - python -m pip install "browsertrace[ui]"
    - export BROWSERTRACE_HOME="$CI_PROJECT_DIR/.browsertrace"
    - RUN_ID=$(browsertrace demo | awk -F': ' '/Run ID:/ {print \$2}')
    - browsertrace export "$RUN_ID" --public -o public.html
  artifacts:
    when: always
    paths:
      - public.html