Show HN: Argus, agentic QA for teams whose coding agents move faster than QA

Aug 19, 2026 02:10 AM - 2 hours ago 1

AI agents that trial your UI for illustration a existent personification — nary scripts to write, nary selectors to maintain.

Website

Argus is simply a ocular UI testing agent. Describe a test, constituent it astatine an HTTP(S) page, and watch it inspect the page successful an isolated Playwright browser context. Runs, timelines, screenshot references, and system reports are captured automatically.

Argus finds the bugs you didn't constitute tests for. Point it astatine a page, picture what "working" looks like, and an autonomous supplier explores your UI the measurement a existent personification would — clicking, typing, scrolling — past hands you a system study pinch screenshots and a timeline. No trial scripts to maintain, nary flaky selectors to babysit.

Argus dashboard

Agentic, not scripted — the supplier reasons astir the page and adapts, it doesn't replay a fixed script
Fits your stack — built connected Playwright, useful against localhost and private-network apps
Zero setup ceremony — commencement testing successful minutes

The fastest measurement to tally Argus: nary install, nary API keys to manage, thing to self-host.

Get started astatine argustest.com.

A tally isn't a azygous exemplary telephone — it's 5 agents handing disconnected to each other, and you tin watch each 1 activity successful the unrecorded tally view:

Validator — checks the target URL and trial explanation are really testable earlier a tally starts.
Comprehender — sounds the trial explanation and breaks it into chopped trial cases.
Explorer — crawls the app first, mapping retired pages and the actions disposable connected each.
Strategist — turns the representation and trial cases into a actual step-by-step plan.
Executor — runs the scheme successful a existent browser: navigating, typing, clicking, and confirming outcomes arsenic it goes.

Nothing happens successful a achromatic container — each supplier streams its reasoning arsenic it works, down to individual actions for illustration "Navigating to /companies" aliases "Confirming 'Airbnb' is connected the page," truthful you spot the app get mapped, the scheme get built, and the trial get executed, live.

Describing a test

A tally successful progress

A completed tally report

Integrations

A trial is conscionable a explanation of intent — Argus figures retired really to interact pinch the page. For example, pointed astatine Y Combinator's site:

Target: https://www.ycombinator.com Test the startup directory. 1. From the homepage, navigate to the companies/startup directory. 2. Search for a well-known YC institution by sanction (e.g. "Airbnb") and corroborate it appears successful the results. 3. Filter the directory by a circumstantial batch (e.g. "Winter 2024") and corroborate the listed companies update to match. 4. Open a company's floor plan from the results and corroborate its name, one-line description, batch, and website nexus each render correctly. 5. Navigate backmost to the directory and corroborate the search/filter authorities behaves arsenic expected — either preserved aliases reset, whichever the page is designed to do. 6. Resize the viewport to 375px width and corroborate the nav collapses into a mobile menu, and the directory database stays scrollable and usable pinch nary overlapping elements. Fail the trial if the known institution doesn't look successful hunt results, if the batch select doesn't really select the list, if a institution floor plan is missing expected fields, aliases if the mobile layout breaks.

Argus runs this for illustration a personification would — clicking done the flow, reference the page to judge occurrence aliases nonaccomplishment — and returns a timeline, screenshots astatine each step, and a pass/fail study pinch the reasoning down it.

Argus is source-available and afloat local-first — SQLite storage, nary telemetry, your information and screenshots ne'er time off your machine. Prefer to self-host? Follow the steps below.

Python 3.11+, uv, and Node.js 20.19+ aliases 22.12+
A Gemini API key for existent runs

cp .env.example .env # Set GEMINI_API_KEY successful .env uv sync --dev uv tally playwright instal chromium cd frontend && npm instal && npm tally build && cd .. uv tally uvicorn argus.app:app --reload --env-file .env

Open http://localhost:8000. For frontend basking reload, tally npm tally dev successful frontend/ alongside Uvicorn and unfastened http://localhost:5173.

Configuration is environment-only:

Variable Default Purpose

GEMINI_API_KEY	—	Required for existent execution
GEMINI_MODEL	gemini-2.5-flash	Gemini REST model
ARGUS_DATA_DIR	data	SQLite and screenshot directory
ARGUS_HEADLESS	true	Playwright browser mode
ARGUS_RUN_TIMEOUT	300	Run timeout successful seconds

Argus accepts normal HTTP(S) targets, including trusted localhost and private-network apps. It rejects credentials and delicate query parameters successful target URLs, and ne'er stores supplier secrets, typed browser values, aliases inspected page content. The settings surface only shows whether supplier configuration is present.

cp .env.example .env # Set GEMINI_API_KEY successful .env docker constitute up --build

The UI is disposable astatine http://localhost:8000 and persistent information is written to ./data.

uv tally pytest uv tally ruff cheque argus tests uv tally pyright argus tests cd frontend && npm tally typecheck && npm tally lint && npm tally build

argus/runtime: provider-neutral agent, message, tool, and convention boundary
argus/providers/gemini.py: earthy httpx Gemini REST/SSE adapter (no supplier SDK)
argus/pipeline.py: planning, 1 browser agent, grounds capture, and reporting
argus/store.py: SQLite runs, events, screenshots, and reports
argus/app.py: REST API, reconnectable per-run WebSockets, and built UI serving
frontend: dashboard/composer, unrecorded session, history, report, and read-only settings

API docs are disposable astatine /docs. All information is local; location is nary authentication aliases multi-user isolation successful this release.

Argus is source-available nether the Argus Source-Available License 1.0 (ASAL-1.0).

Individuals & Small Teams (< 100 members):to use, modify, and self-host for some commercialized and non-commercial purposes.
Enterprises (100+ members): Requires a commercialized license. Contact america astatine [email protected] to get group up.
Releases automatically person to the MIT License aft 3 years.