BrowserTrace v0.1.17: PyPI-published local flight recorder for AI browser agents

[aaronlab]

BrowserTrace v0.1.17 is published on PyPI and ready for first users.

It records AI browser-agent runs as a local timeline: screenshot, URL, action, model input/output, status, and error. The goal is to make failures from Browser Use, Stagehand, Skyvern, Playwright + LLM, and custom computer-use agents debuggable without staring at logs.

Zero-install live demo:
https://aaronlab.github.io/browsertrace/

Try it locally in 60 seconds 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

Persistent install:

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

Then open http://127.0.0.1:3000 and inspect the failed checkout-agent run. browsertrace doctor works before any trace database exists, so it is the first command to run if setup fails.

Public-safe export:

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

--public omits prompt/model I/O, screenshots, and URLs from the standalone HTML export. Individual redaction flags remain available when you want to keep some fields visible.

Useful links:

• PyPI: https://pypi.org/project/browsertrace/
• Release: https://github.com/aaronlab/browsertrace/releases/tag/v0.1.17
• Live demo: https://aaronlab.github.io/browsertrace/
• Public-safe demo export: https://github.com/aaronlab/browsertrace/releases/download/v0.1.17/browsertrace-demo-public.html
• Walkthrough: https://aaronlab.github.io/browsertrace/debug-browser-agent-failure.html
• 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
• Comparison: https://aaronlab.github.io/browsertrace/compare-browser-agent-debugging.html
• Launch feedback: Launch feedback: what browser-agent failures should BrowserTrace capture? #3
• Good first issues: good first issue Good for newcomers
• First PR Recipe: https://github.com/aaronlab/browsertrace/blob/main/CONTRIBUTING.md#first-pr-recipe

Feedback wanted:

• What browser-agent failure was hardest for you to debug?
• Which stack should BrowserTrace support better first: Browser Use, Stagehand, Skyvern, Playwright, or custom computer use?
• Which artifacts should be captured next: DOM/extracted text, selectors, retries, workflow state, model input/output, or something else?
• Is a portable HTML export enough for sharing failures, or do you need hosted links?

Please do not post private prompts, screenshots, credentials, cookies, tokens, customer data, or private trace data in public issues/discussions. Use browsertrace export --public before sharing a real trace publicly, and use the Security Policy for security-sensitive reports or changes.

No star asks, upvote asks, vote swaps, or artificial engagement. Workflow feedback from real browser-agent builders is the useful thing.

[aaronlab]

Roadmap tracking is now open here: https://github.com/aaronlab/browsertrace/milestone/1

Current v0.2 launch follow-up items:

• PyPI publishing
• First-class Skyvern adapter
• Launch feedback from browser-agent builders

[aaronlab]

Added a value-first walkthrough for people who want the debugging workflow before trying the repo:

https://aaronlab.github.io/browsertrace/debug-browser-agent-failure.html

It shows the failure pattern, the local trace commands, and the questions to ask at the first red step.

[aaronlab]

Integration feedback is now structured.

If you are using Browser Use, Stagehand, Playwright + LLM scripts, Skyvern, or a custom computer-use agent, please open an integration request here:

https://github.com/aaronlab/browsertrace/issues/new?template=integration_request.yml

The form asks which stack you use, what failure current logs do not explain, which hooks BrowserTrace should wrap, and whether you can test a preview adapter.

[aaronlab]

Launch update: the public sharing assets are now ready.

Useful links:

• Live demo: https://aaronlab.github.io/browsertrace/
• Launch kit: https://aaronlab.github.io/browsertrace/launch/
• Debugging walkthrough: https://aaronlab.github.io/browsertrace/debug-browser-agent-failure.html
• Repo: https://github.com/aaronlab/browsertrace

Prepared launch materials now include Day 1 social posts, Show HN notes, targeted Browser Use / Stagehand / Playwright / Skyvern community drafts, Product Hunt listing copy, and a Chinese long-form technical post.

The most useful feedback is still workflow-specific: if you build browser agents, what state do you wish your trace captured at failure time? Screenshots, DOM/extracted text, selectors, model input/output, workflow state, retries, or something else?

[aaronlab]

Release update: BrowserTrace v0.1.2 is now available.

Release:
https://github.com/aaronlab/browsertrace/releases/tag/v0.1.2

What changed:

• Added a basic Skyvern tracing wrapper: browsertrace.integrations.skyvern.wrap_skyvern
• Records high-level Skyvern-shaped run_task and run_workflow calls
• Supports async and sync client methods
• Captures method input, returned run IDs/status, and step-level errors
• Added public Run.close(error=None) so integration-created runs can be marked completed or failed through the documented path
• Updated the public launch kit and release assets

The next useful feedback is still workflow-specific: for Browser Use, Stagehand, Skyvern, Playwright + LLM, or custom computer-use agents, which failure artifacts should BrowserTrace capture first?

[aaronlab]

Release update: BrowserTrace v0.1.7 is now available.

Release:
https://github.com/aaronlab/browsertrace/releases/tag/v0.1.7

New since the earlier launch discussion:

• No-dependency Browser Use-shaped callback demo: examples/browser_use_callback_demo.py
• Browser-agent debugging comparison page: https://aaronlab.github.io/browsertrace/compare-browser-agent-debugging.html
• Playwright + LLM guide: https://aaronlab.github.io/browsertrace/playwright-llm-debugging.html
• Playwright + LLM feedback issue: Feedback wanted: Playwright + LLM trace fields #12
• Directory/newsletter submission sheet: https://aaronlab.github.io/browsertrace/launch/directory-submission-sheet.md

The most useful feedback is still concrete failure context: for Browser Use, Stagehand, Skyvern, Playwright + LLM, or custom computer-use agents, what did your logs or traces miss when the run failed?

[aaronlab]

Release update: BrowserTrace v0.1.8 is now available.

Release:
https://github.com/aaronlab/browsertrace/releases/tag/v0.1.8

New in this release:

• browsertrace export --redact-screenshots for HTML exports where screenshots may reveal private page data
• Sharing docs now distinguish prompt/model I/O redaction from screenshot redaction

Public sharing commands:

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

Feedback wanted: when you share browser-agent traces publicly, what else needs redaction besides prompt/model I/O and screenshots?

[aaronlab]

Release update: BrowserTrace v0.1.9 is now available.

Release:
https://github.com/aaronlab/browsertrace/releases/tag/v0.1.9

New in this release:

• browsertrace export --redact-urls for HTML exports where step URLs may reveal private domains, query strings, tokens, or internal paths
• Public sharing docs now show prompt/model I/O, screenshot, and URL redaction together

Public sharing command:

browsertrace export <run_id> --redact --redact-screenshots --redact-urls -o public.html

Feedback wanted: what else should a browser-agent trace be able to redact before sharing publicly?

[aaronlab]

Launch kit update for v0.1.10:

• Public-safe demo export is now linked across the README, launch kit, Day 1-4 packets, directory sheet, and long-form tutorial drafts.
• Downloadable public-safe asset: https://github.com/aaronlab/browsertrace/releases/download/v0.1.10/browsertrace-demo-public.html
• scripts/launch_metrics.py now records GitHub traffic views/clones in addition to stars, forks, issues, and release downloads.
• GitHub profile README draft for the renamed aaronlab account is prepared in docs/launch/github-profile-readme.md; owner action is tracked in Owner action: update aaronlab GitHub profile README #13.

Current launch stance is unchanged: ask for real browser-agent debugging workflow feedback, not stars, upvotes, or vote swaps.

[aaronlab]

Launch log update:

• Added owner-action blocker Owner action: upload GitHub social preview image #15 for uploading the prepared GitHub social preview image: Owner action: upload GitHub social preview image #15
• Updated the shortest owner checklist and publish queue so the social preview upload happens before external posts.
• Added the canonical URL to the main debugging tutorial and verified GitHub Pages deployed it.
• Current live repo audit remains 3 stars / 1 fork / 0 watchers, so the 1000-star goal is still incomplete.

No paid stars, fake engagement, vote swaps, or direct star asks.

[aaronlab]

Launch log update:

• Added owner-action issue Owner action: submit GitHub Pages sitemap to search consoles #16 for submitting the GitHub Pages sitemap to Google Search Console and Bing Webmaster Tools: Owner action: submit GitHub Pages sitemap to search consoles #16
• Added the owner checklist at docs/launch/search-indexing-submission.md and exposed it from the launch kit.
• Verified the published checklist is live at https://aaronlab.github.io/browsertrace/launch/search-indexing-submission.md
• Current live repo audit remains 3 stars / 1 fork / 0 watchers, so the 1000-star goal is still incomplete.

Search indexing is a long-tail discovery channel only. No fake traffic, click farms, doorway pages, or keyword stuffing.

[aaronlab]

Launch log update:

PyPI is still the main install-conversion blocker, but the release now has a verified no-install uvx trial path from the GitHub tag:

uvx --from "browsertrace[ui] @ git+https://github.com/aaronlab/browsertrace@v0.1.10" browsertrace demo
uvx --from "browsertrace[ui] @ git+https://github.com/aaronlab/browsertrace@v0.1.10" browsertrace list
uvx --from "browsertrace[ui] @ git+https://github.com/aaronlab/browsertrace@v0.1.10" browsertrace

Verified locally with an isolated BROWSERTRACE_HOME: the demo run is created and browsertrace list shows the failed checkout-agent run.

This is only a bridge until PyPI Trusted Publisher is configured. The launch goal remains unchanged: get real feedback from people debugging Browser Use, Stagehand, Skyvern, Playwright + LLM, and custom computer-use agent failures. No paid stars, fake engagement, vote swaps, or direct star asks.

[aaronlab]

v0.1.11 is now published and verified.

The main first-run addition is browsertrace doctor, which works before any trace database exists:

uvx --from "browsertrace[ui] @ git+https://github.com/aaronlab/browsertrace@v0.1.11" browsertrace doctor
uvx --from "browsertrace[ui] @ git+https://github.com/aaronlab/browsertrace@v0.1.11" browsertrace demo
uvx --from "browsertrace[ui] @ git+https://github.com/aaronlab/browsertrace@v0.1.11" browsertrace list
uvx --from "browsertrace[ui] @ git+https://github.com/aaronlab/browsertrace@v0.1.11" browsertrace

Verified from the GitHub tag with isolated BROWSERTRACE_HOME. Release assets include wheel, sdist, full demo HTML, public-safe demo HTML, demo video, poster, and GIF.

PyPI is still pending owner-side Trusted Publisher setup, so GitHub install/uvx remains the public trial path for now.

[aaronlab]

Computer-use update for custom browser-agent builders:

• New guide: https://aaronlab.github.io/browsertrace/computer-use-agent-debugging.html
• New runnable no-browser example: python examples/computer_use_loop_example.py
• Examples matrix: https://github.com/aaronlab/browsertrace/tree/main/examples

The example records a generic observe -> decide -> act loop, then marks the model-selected checkout click as the failed step. I am especially looking for feedback on which fields custom computer-use agents should capture by default: observation text, screenshot, URL, action label, model input/output, selector, retries, latency, or redaction policy.

[aaronlab]

Launch/contributor update 2026-05-09:

• Merged the first follow-up PR from an external contributor: docs: add uvx troubleshooting note #22 by @nightcityblade. It adds a short uvx: command not found troubleshooting note while keeping the README on the verified v0.1.12 install links.
• Opened the next good first issue: Docs: add browsertrace doctor troubleshooting notes #23 for concise browsertrace doctor troubleshooting notes.
• Current live audit remains 3 stars / 2 forks, so the 1000-star goal is still incomplete.

Still asking for browser-agent debugging workflow feedback only. No star asks, upvote asks, vote swaps, or artificial engagement.

[aaronlab]

Launch/conversion update 2026-05-09:

• Updated the main discussion good-first link to point at Docs: add browsertrace doctor troubleshooting notes #23, the current browsertrace doctor troubleshooting issue.
• Added the verified doctor -> demo first-run path across the remaining launch copy, long-form posts, response templates, and the live aaronlab profile README.
• Added PyPI project links for the debugging guide, computer-use guide, and Browser Use guide so the package page is more useful once Trusted Publisher is configured.
• Current audit remains 3 stars / 2 forks / 0 watchers; PyPI still returns 404 until owner-side publishing is configured.

Still looking for real browser-agent debugging feedback: which stack fails in ways your logs cannot explain, and what state should BrowserTrace capture at the failed step?

[aaronlab]

Contributor update:

• Update README.md #27 merged the examples README public-safe export flow.
• Docs: expand examples README public-safe export flow #26 is closed.
• The current good first issue is Docs: add Windows public-safe export flow to examples README #28: add the Windows PowerShell variant for the same public-safe export flow.
• Latest local verification: uv run --python 3.11 --extra dev pytest -q -> 94 passed.

Current trial path remains the v0.1.13 GitHub-tag uvx flow until PyPI Trusted Publisher is configured.

[aaronlab]

v0.1.14 is published: https://github.com/aaronlab/browsertrace/releases/tag/v0.1.14

What changed since v0.1.13:

• Browser Use adapter records compact browser-state context in model_input.
• Browser Use guide documents callback compatibility and troubleshooting.
• Stagehand wrapper records successful act, extract, observe, goto, and click return values as model_output.
• PR template and security reporting path are clearer for contributors.

Verified local trial path:

uvx --from "browsertrace[ui] @ git+https://github.com/aaronlab/browsertrace@v0.1.14" browsertrace --help

PyPI is still owner-blocked, so public install copy should keep using the GitHub tag until Trusted Publisher is configured.

[aaronlab]

Launch/contributor update:

• Added examples/playwright_llm_loop_example.py, a no-network Playwright + LLM-shaped trace for prompt/messages, DOM/accessibility snippets, selector/action, retry metadata, and a failed step.
• Added the Windows PowerShell public-safe export flow to examples/README.md and closed Docs: add Windows public-safe export flow to examples README #28.
• Rotated the current good-first contribution path to Docs: add pytest fixture recipe for isolated trace storage #29: pytest fixture recipe for isolated trace storage.
• Refreshed llms.txt, press kit, profile draft, and live aaronlab profile README to point to Docs: add pytest fixture recipe for isolated trace storage #29.
• Submitted updated docs/launch URLs to IndexNow; response was HTTP 200.

Latest verified local suite: uv run --python 3.11 --extra dev pytest -q -> 104 passed.

Still looking for workflow feedback from people debugging Browser Use, Stagehand, Skyvern, Playwright + LLM, or custom computer-use agents. The most useful feedback is which browser state or model decision context was missing when a real run failed.

[aaronlab]

Launch/contributor update 2026-05-10:

• README first-run notes now cover the full local path: browsertrace doctor, browsertrace demo, browsertrace list, browsertrace show, public-safe export, run ID output, export filenames, and SDK-only terminal commands.
• Current good first issue: Docs: add README export success output note near install checks #119, a small README note for the browsertrace export success output.
• Directory discovery: BrowserTrace has an open agentfirst.directory PR under browser interaction tools: Add BrowserTrace to browser interaction tools bradvin/agentfirst.directory#30
• Latest local verification: uv run --python 3.11 --extra dev pytest -q -> 195 passed. Latest main CI and Pages are green.

Still looking for concrete workflow feedback from people debugging Browser Use, Stagehand, Skyvern, Playwright + LLM, or custom computer-use agents: what browser state or model-decision context did your logs miss at the failed step?

[aaronlab]

Launch/discovery update 2026-05-10:

High-fit external discovery PRs currently open for BrowserTrace:

• Add BrowserTrace to browser interaction tools bradvin/agentfirst.directory#30
• Add BrowserTrace angrykoala/awesome-browser-automation#112
• Add BrowserTrace mxschmitt/awesome-playwright#136
• Add BrowserTrace Jenqyang/Awesome-AI-Agents#220
• Add BrowserTrace wjhou/awesome-computer-use-agents#2
• Add BrowserTrace cdxeve/awesome-computer-use-agents#2
• Add: BrowserTrace steel-dev/awesome-web-agents#56
• Add BrowserTrace to Debugging & Developer Experience ai-boost/awesome-harness-engineering#23

These are framed as browser-agent debugging / observability / harness-engineering entries, not as generic AI promotion. Current status: open, no maintainer change requests from us to handle right now.

Current small contribution path: #223.

Current repo audit: 3 stars / 5 forks / 0 watchers, so the 1000-star goal is still incomplete. Remaining owner-side conversion blockers are PyPI Trusted Publisher, GitHub social preview upload, and pinning BrowserTrace on the aaronlab profile.

No paid stars, fake engagement, vote swaps, star asks, or upvote asks. The useful ask remains workflow feedback from people debugging Browser Use, Stagehand, Skyvern, Playwright + LLM, or custom computer-use agent failures.

[aaronlab]

Launch/discovery status update (2026-05-10):

• Current good-first contribution path is now Docs: link Security Policy from directory troubleshooting reply #225, and it is pinned with Launch feedback: what browser-agent failures should BrowserTrace capture? #3 and Integration feedback: Browser Use adapter capture fields #11. Closed Docs: link Security Policy from awesome-list troubleshooting reply #223 was unpinned so new contributors do not land on a completed task.
• BrowserTrace repo metadata is set: homepage, description, MIT license, and AI/browser-agent topics are live.
• External discovery PRs remain open/mergeable with no maintainer feedback requiring a response right now.
• The E2B awesome SDKs PR is the only known blocked external PR; it needs the owner-side CLA signature before verification/cla-signed can pass.
• IndexNow was refreshed after the Docs: link Security Policy from directory troubleshooting reply #225 rotation; generic and Bing endpoints returned HTTP 200.

Current repo status from GitHub: 3 stars, 5 forks. Still far from the 1000-star goal, so the next work remains legitimate discovery, useful docs, contributor care, and owner-only launch actions. No fake engagement, vote swaps, or star asks.

[aaronlab]

v0.1.15 is now published with Browser Use run-hook support.

Release: https://github.com/aaronlab/browsertrace/releases/tag/v0.1.15
Browser Use guide: https://aaronlab.github.io/browsertrace/browser-use-debugging.html

What changed:

• Added create_run_hooks(tracer, name=...) for Browser Use apps that pass on_step_start and on_step_end directly to agent.run(...).
• Kept attach_tracer(agent, ...) for register_new_step_callback and compatible callback-attribute agents.
• Refreshed README, docs, launch kit, llms.txt, and release assets to the pinned v0.1.15 tag.

Verification:

• Local full suite: 299 passed.
• Local build and twine check: passed for wheel and sdist.
• Remote CI on main: success for Python 3.11, 3.12, 3.13, plus build.
• Remote Pages: success.
• Verified GitHub-tag trial path: uvx --from "browsertrace[ui] @ git+https://github.com/aaronlab/browsertrace@v0.1.15" browsertrace doctor/demo/list works.
• Verified tag install exposes create_run_hooks.

No fake engagement or star asks. Current growth blockers remain owner-only: PyPI Trusted Publisher, GitHub social preview upload, profile pin, E2B CLA, and real owner social/community posts.

[aaronlab]

Historical note (2026-05-11): this comment originally announced the first PyPI release. BrowserTrace is now current at browsertrace==0.1.17.

Current quick trial path:

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

Persistent install:

pip install "browsertrace[ui]"

Release: https://github.com/aaronlab/browsertrace/releases/tag/v0.1.17
Public-safe demo export: https://github.com/aaronlab/browsertrace/releases/download/v0.1.17/browsertrace-demo-public.html

The main discussion body has been refreshed to the current PyPI install path and release assets. Still looking for workflow feedback from people debugging Browser Use, Stagehand, Skyvern, Playwright + LLM, or custom computer-use failures. No star/upvote asks.

[aaronlab]

Contributor queue update (2026-05-11):

• Docs: add expected-output notes for no-service examples #234 is claimed by a contributor for the no-service example expected-output docs task. The claimed label is present so others do not duplicate that work.
• Docs: add public-safe export verification note #242 was completed by contributor PR docs: add public export verification note #243.
• Docs: add expected-output note for isolated pytest trace store #244 was completed by contributor PR docs: clarify isolated pytest trace-store output #245.
• Docs: add expected-output note for JSON CLI checks #246 was completed by contributor PR docs: add expected JSON CLI output notes #247.
• Docs: add environment variable example values #248 is the current unclaimed pinned good-first issue: add a small environment-variable expected-values note in examples/README.md.
• Current pinned issues: Launch feedback: what browser-agent failures should BrowserTrace capture? #3 for launch feedback, Integration feedback: Browser Use adapter capture fields #11 for Browser Use adapter feedback, and Docs: add environment variable example values #248 for an open first contribution path.

If you want to help, #248 is intentionally docs-only and narrow. Please include the exact commands you ran when opening a PR. No stars/upvotes requested; concrete workflow and docs feedback is the useful signal.