What is Sentri?

Sentri is an autonomous QA platform that crawls your web app, generates Playwright test suites using AI, and self-heals them when your UI changes.

The Problem

Problem	How Sentri solves it
Writing E2E tests is slow	Point it at a URL — tests are generated in minutes, not days
Selectors break every sprint	Self-healing runtime tries multiple strategies, remembers what worked
AI-generated tests are untrustworthy	Every test lands in a Draft queue. Nothing executes until a human approves
You can't see what the test is doing	Live browser screencast, real-time log stream, per-step screenshots
Tests fail and nobody knows why	AI classifies every failure and auto-regenerates the worst offenders
Vendor lock-in on AI providers	Swap between Anthropic, OpenAI, Google, or Ollama with one setting

How It Works

1. Crawl

Sentri launches a real Chromium browser and explores your app — following links, mapping forms, buttons, and interactive elements. A D3 force-directed Site Graph shows discovered pages in real time. While crawling, Sentri captures every API call (fetch/XHR) your app makes — building a map of backend endpoints automatically.

2. Generate

Each page goes through an 8-stage AI pipeline: crawl → filter → classify → plan → generate → deduplicate → enhance → validate. All tests land in a Draft queue.

In addition to UI tests, Sentri generates API contract tests in two ways:

• Automatic: During crawl, HAR capture records every fetch/XHR call and feeds endpoints to the AI
• From description: In the Generate Test modal, describe your API endpoints in plain English (e.g. write API tests for https://reqres.in/api/register), paste GET /api/users patterns with request/response examples, or attach an OpenAPI spec as a .json file

API tests use Playwright's request API context to call endpoints directly — no browser needed. See the API Testing guide for example prompts and OpenAPI spec support.

3. Review

Approve or reject tests one by one or in bulk. Only approved tests execute in regression.

4. Execute

One-click regression with live browser view (CDP screencast), SSE log stream, execution timeline, and per-step screenshots with bounding-box overlays. Run up to 10 tests in parallel — each in its own isolated browser context with independent video, screenshots, and network logs. Select ⚡ 4x from the project action bar or set PARALLEL_WORKERS=4 in your .env.

5. Self-Heal

When a selector fails, the self-healing layer tries fallback strategies in a waterfall. When a fallback wins, it records which strategy succeeded — and tries it first next time.

6. Monitor

Dashboard with pass rate, defect categories, flaky test detection, test growth sparkline, and self-healing stats.

Architecture

┌─────────────────────────────────────────────┐
│  Frontend (React + Vite)                    │
│  ├── Pages: Dashboard, Projects, Tests, ... │
│  ├── SSE: real-time run events              │
│  └── CDP: live browser screencast           │
├─────────────────────────────────────────────┤
│  Backend (Express + Node.js)                │
│  ├── Routes: /api/v1/projects, /tests, /runs │
│  ├── Crawler: Playwright + SmartCrawlQueue  │
│  ├── HAR Capture: API traffic → endpoints   │
│  ├── AI Pipeline: 8-stage generation        │
│  ├── Self-Healing: multi-strategy waterfall │
│  ├── DB: SQLite (default) or PostgreSQL     │
│  └── Redis (optional): rate limits, SSE     │
├─────────────────────────────────────────────┤
│  AI Providers                               │
│  ├── Anthropic Claude                       │
│  ├── OpenAI GPT-4o-mini                     │
│  ├── Google Gemini                          │
│  └── Ollama (local, free)                   │
└─────────────────────────────────────────────┘