Skip to main content
Spawn parallel AI agents that navigate a live web application using a real browser, find bugs, take screenshots, and optionally create GitHub issues or PRs with their findings. 
paragon autotest [description] [flags]
Unlike paragon test which runs pre-written Playwright tests, autotest uses AI agents that explore your app autonomously — no test files needed. Each agent gets its own isolated browser instance.

​
How It Works

  1. Planning — A planning agent analyzes the target app and distributes testing work across agents.
  2. Exploration — Each agent navigates the app in a real browser, performing actions like clicking, typing, scrolling, and navigating between pages.
  3. Reporting — Findings are compiled into a report with screenshots, reproduction steps, and severity ratings. Playwright test code is auto-generated from the agent’s actions.
  4. GitHub integration (optional) — Findings are filed as GitHub issues or bundled into a PR.

​
Flags

FlagTypeDefaultDescription
--urlstring(required)Target URL to test.
--promptstring—What to test (e.g. "test the checkout flow").
--agentsint1Number of parallel agents (1–8). Each agent gets its own browser.
--modelstringparagon-maxModel to use for the agents.
--timeoutint15Per-agent timeout in minutes.
--authstring—Path to a cookies JSON file (browser-use export format) for pre-authenticated sessions.
--loginstring—Login instructions for the agent in plain English (e.g. "click Login, type user@test.com into email, type pass123 into password, click Submit").
--sessionstring—Team auth session name or ID stored in Supabase.
--repostring—GitHub repo for issue/PR creation (owner/repo).
--prboolfalseCreate a PR with findings instead of an issue. Defaults to true when --repo is set.
--skip-githubboolfalseSkip GitHub issue/PR creation even if --repo is set.
--jsonboolfalseOutput results as structured JSON.
-q, --quietboolfalseMinimal output.

​
Examples

# Basic: test a local app
paragon autotest --url http://localhost:3000 --prompt "make sure the page loads"

# Run against production with multiple agents
paragon autotest --url https://yourapp.com --prompt "test the checkout flow" --agents 4

# Test with login instructions
paragon autotest --url https://yourapp.com --login "click Login, type user@test.com into email, type pass123 into password, click Submit" --prompt "test the dashboard"

# Use a cookies file for auth
paragon autotest --url https://yourapp.com --auth ./cookies.json --prompt "test account settings"

# Use a shared team auth session
paragon autotest --url https://yourapp.com --session "staging-admin" --prompt "test admin panel"

# File findings as a GitHub PR
paragon autotest --url https://yourapp.com --prompt "test user registration" --repo acme/webapp --pr

# Skip GitHub, just see findings locally
paragon autotest --url http://localhost:3000 --prompt "test login" --skip-github

# JSON output for CI pipelines
paragon autotest --url https://yourapp.com --prompt "smoke test" --json --quiet

​
Authentication

Autotest supports three ways to authenticate agents with your app:
MethodFlagWhen to use
Login instructions--loginSimple login forms. Describe the steps in plain English.
Cookies file--authExport cookies from browser-use and pass the JSON file.
Team session--sessionShared, encrypted auth sessions stored in Supabase for team reuse.

​
Managing Team Auth Sessions

Use paragon autotest auth to manage shared authentication sessions: 
# List all team auth sessions
paragon autotest auth list

# Capture a new session (opens a browser — log in manually, then close)
paragon autotest auth capture https://yourapp.com --name "staging-admin"

# Use the captured session
paragon autotest --url https://yourapp.com --session "staging-admin" --prompt "test dashboard"

# Delete a session
paragon autotest auth delete "staging-admin"

​
Output

Each finding includes:
  • Title and description of the bug
  • Severity rating (critical, high, medium, low)
  • Screenshots captured during the session
  • Reproduction steps from the agent’s browser actions
  • URL where the issue was found
  • Auto-generated Playwright code to reproduce the bug
When --repo is set, findings are automatically filed as GitHub issues or bundled into a PR with the generated Playwright tests.