> ## Documentation Index
> Fetch the complete documentation index at: https://docs.argalabs.com/llms.txt
> Use this file to discover all available pages before exploring further.

# How Arga works

> Natural language tests, deterministic sandboxes, auto-heal

Arga turns natural language into reliable, production-grade tests. Describe what to test in plain English. Arga compiles it into a deterministic workflow, runs it in a fresh sandbox with API twins, and auto-heals when things break instead of just failing.

## Two ways to validate

<Steps>
  <Step title="Point at your staging and let Arga test">
    Give Arga any URL — staging, preview, or localhost. It generates and runs tests against your app with API twins standing in for real services. No infra to set up.
  </Step>

  <Step title="Deploy with Arga and get sandboxed testing">
    Connect your GitHub repo. On every PR, Arga deploys your changed services into a fresh sandbox with API twins and runs tests from the diff. Results post back as a GitHub check.
  </Step>
</Steps>

## Under the hood

Arga is built on three primitives:

<CardGroup cols={3}>
  <Card title="Semantic abstraction" icon="layer-group">
    Tests target elements by intent and meaning, not CSS selectors or XPaths. UI changes don't break tests unless behavior actually changes. Service fork setup requires a one-time onboarding call — [book a call](https://cal.com/akiratong/30min?overlayCalendar=true) so we can understand your stack.
  </Card>

  <Card title="Deterministic sandboxes" icon="clone" href="/concepts/digital-twins">
    Each run starts from a fresh, identical state with API twins (Stripe, Slack, and more). Same input, same output, every time.
  </Card>

  <Card title="Auto-heal" icon="rotate" href="/concepts/session-replay">
    When something fails, Arga reads from the browser and gathers context to continue instead of just failing.
  </Card>
</CardGroup>

## Context from the tools you ship with

Arga pulls context from [integrations](/integrations) like Jira, Linear, GitHub, Sentry, and PostHog, so the agents generating tests know what to validate and why it matters on each change. No fixtures to maintain, no test plans to keep up to date.

## Beyond code: validating AI agents

The same infrastructure that validates code changes also validates AI agents. Drop an agent into a sandbox alongside the same digital twins, and you can:

* Observe agent behaviour without real-world side effects
* Proactively red-team agents to discover unsafe or unexpected actions
* Validate that agents interact correctly with external APIs

<Card title="Try it yourself" icon="rocket" href="/quickstart" horizontal>
  See the full pipeline in action — run your first validation in under a minute.
</Card>