render
Open the page in a real browser before judging visual state, layout, or interaction behavior.
native agent browser
Browser evidence for agent work.
Bitter Browser gives automation a real local viewport and a receipt trail: screenshots, DOM snapshots, traces, and repeatable verification commands.
screenshotDOM snapshottracecommand
receipt4 artifacts captured
observefingerprint changed
shotartifacts/frame-001.png
snapsnapshots/page.html
tracetrace.ndjson
what it is for
A browser check should leave a trail that another person can review.
receipt
Preserve the screenshot, snapshot, trace, and command path that prove what the run actually observed.
verify
Separate behavior that was exercised from behavior the agent merely expected to work.
receipt model
Proof is useful only when the next reviewer can replay the path.
Bitter Browser is framed around behavior receipts: compact artifacts that say what URL was opened, what viewport was used, which action was taken, and where the resulting evidence lives. The goal is not to make agents sound certain. The goal is to make their claims inspectable.
What the browser rendered
A visual receipt shows the actual viewport the agent inspected, not a claim inferred from markup or logs.
What the DOM contained
HTML and accessibility state make it possible to compare what changed between attempts without rerunning the session.
What happened in sequence
Trace files tie navigation, inputs, timing, and network observations to one run so reviewers can follow the behavior.
How to repeat the check
Receipts point back to the exact browser command or QA script that produced the evidence.
safe automation
Browser automation should be explicit, local, and reviewable.
- Local-first sessions keep automation close to the worktree and make browser actions explicit.
- Receipts describe page behavior and artifacts; they should never be used to publish credentials or private tokens.
- Verification commands give reviewers a repeatable path before a deploy is trusted.
launch readiness
build
bun run generatebrowser smoke
bun run qa:smokelive verify
verify_deploy bitterbrowser.comThese checks are receipts, not decoration. A launch is healthier when every product claim has a current build, browser, and deploy proof attached to it.
faq
Questions first users ask before trusting a browser receipt.
Who is Bitter Browser for?
It is for agents and reviewers who need browser proof while building, repairing, or verifying web properties.
What counts as a receipt?
A useful receipt names the URL, viewport, action, screenshot, snapshot, trace, and command that produced the evidence.
Does this replace product QA?
No. Bitter Browser complements repo-owned smoke and end-to-end tests by making the browser evidence easier to inspect.