Your product's spec, alive.
I see · I guard · I protect
An autonomous QA agent that tests your web app in a real browser, remembers how your product behaves, and tells you the moment something silently breaks.
npm install -g https://github.com/im-anp/argus-releases/releases/latest/download/argusqa-latest.tgz
The problem
How does this feature behave today?
In most teams the answer lives in one senior QA's head. When they leave, it leaves with them.
It worked last week.
The worst bugs don't crash. The page renders fine while the API behind it returns 500s. Nobody notices until a customer does.
Can you re-check all flows before release?
Someone re-clicks the same fifty flows. Every release. Again.
How it works
What you get
Paste a URL, watch it walk your product live and draft a behavioral spec — pages, flows, invariants.
Every flow classified pass / drift / broken. Drift — "works, but different than recorded" — is the category everyone else misses.
Watches every network call and console error while clicking. UI fine but the backend 500'd? A finding — endpoint, status, and response body quoted.
Exact URL, expected vs observed, numbered repro, API evidence, screenshot. Retest any bug: pass = resolved, later failure = reopened. Automatically.
Every run becomes an execution report — pass rate, per-case results, evidence. Save as PDF or copy as Markdown.
Form logins stored encrypted, replayed by name. SSO/2FA: log in once in a real window, Argus reuses the session. The AI model never sees a credential.
Deleting data, payments, sending messages — Argus asks first and never does them unattended. Every permission visible and revocable.
Drop in a PRD, a mockup, a user's bug screenshot. It drafts test flows, compares design vs reality, reproduces the report.
Product memory
Six months later, nobody remembers why the delete button shows no toast. Argus does — with the evidence still attached.
Integrations
Connect with a browser approval, no tokens to copy. Findings become complete tickets with the screenshot attached — straight from the bug card, always with your say-so.
Nightly watchdog digests land in your channel: what passed, what drifted, what broke — before your standup does.
Trust
100% on your machine. Your app, your data, your history — nothing is sent to us. There is no "us" server at all.
Bring your own key — Anthropic, OpenAI, or Google. Usage bills to your account. Switch anytime with argus provider.
Plain markdown on your disk, built on Google's open OKF knowledge standard. Read it, version it, take it anywhere.
Get started
# Node 24+ required — make sure it's active when installing nvm use 24 npm install -g https://github.com/im-anp/argus-releases/releases/latest/download/argusqa-latest.tgz argus # picks your AI provider, opens http://localhost:3123
argusstart Argus — agent + dashboard at localhost:3123argus providerswitch AI provider or key (Claude / GPT / Gemini)argus updateupdate to the latest versionFAQ
No. Argus runs locally and talks only to the AI provider you chose, under your key and their terms.
No. Argus tests your product the way a user experiences it — through the browser. Production, staging, localhost, or behind your VPN.
Test tools run scripts you write and maintain. Argus builds and maintains its own understanding of your product, verifies it against reality, and answers questions about it. The living spec is the product — tests are a byproduct.
Argus itself is free to install. You pay your AI provider directly for usage, on your own key.