Home

Sentri

Sentri

AI-powered end-to-end test generation, execution, and self-healing for modern web applications.

CI Latest Release Node.js 20+ Playwright MIT License

Get Started  ·  Documentation  ·  API Reference  ·  Roadmap  ·  Changelog

What is Sentri?

Sentri is an autonomous QA platform that covers the full testing lifecycle in a single tool. Point it at a URL — it crawls your application, runs an 8-stage AI pipeline to generate a Playwright test suite, routes tests through a review queue (with optional confidence-based auto-approval for high-quality output), executes approved tests in real browsers across Chromium, Firefox, and WebKit, and automatically repairs broken selectors between runs.

Crawl → Generate → Deduplicate → Enhance → Validate → Review → Execute → Self-Heal

Most AI test generators stop at code generation. Sentri treats generation as step two of eight.

Why Sentri?

Problem How Sentri addresses it
Writing E2E tests is slow Point it at a URL — tests are generated in minutes
Selectors break every sprint Adaptive selector waterfall records what works and tries it first next run
AI-generated tests are untrustworthy Tests land in a Draft queue by default; opt-in confidence-based auto-approval with a full audit trail and one-click revoke for high-quality output
Tests fail and nobody knows why AI feedback loop classifies every failure and auto-regenerates failing tests
No visibility into what the test is doing Live browser screencast, real-time SSE log stream, per-step screenshots
Vendor lock-in on AI providers Switch between Anthropic, OpenAI, Google, OpenRouter, or Ollama with a single setting

Key Features

Test Generation

  • Two discovery modes: Link Crawl maps <a> tags; State Exploration clicks, fills, and submits to discover multi-step flows
  • 8-stage AI pipeline with intent classification, deduplication, assertion enhancement, and structural validation
  • API test generation — captures fetch/XHR traffic during crawl and produces Playwright request contract tests alongside UI tests
  • Natural-language test creation — describe a scenario and skip the crawl entirely

Execution & Observability

  • Parallel execution across 1–10 isolated browser contexts
  • Cross-browser support: Chromium, Firefox, and WebKit with per-run engine selection
  • Live browser screencast at ~7 FPS via Chrome DevTools Protocol
  • Real-time log and result streaming via Server-Sent Events

Self-Healing

  • Multi-strategy selector waterfall: ARIA role → label → text → aria-label → title → CSS
  • Adaptive memory — records the winning strategy per element and prioritises it on subsequent runs
  • Failure classification by category (selector / timeout / assertion / navigation) with targeted regeneration

Operations

  • Flaky test detection with 0–100 scoring based on run history
  • Scheduled runs with timezone support
  • CI/CD webhook trigger with per-project Bearer tokens
  • Diff-aware crawling — only regenerates tests for pages whose DOM fingerprint changed since the last crawl; zero LLM calls when nothing changed
  • Vercel + Netlify deployment webhooks — HMAC-signed payloads auto-launch a diff-aware crawl against the preview URL when a deployment is READY
  • Failure notifications via Microsoft Teams, email, and generic webhook
  • Workspace isolation and role-based access control (Admin / QA Lead / Viewer)
  • GDPR/CCPA account export and cascade deletion

Quick Start

git clone https://github.com/RameshBabuPrudhvi/sentri.git
cd sentri

cp backend/.env.example backend/.env
# Add at least one AI provider key to backend/.env

docker compose up --build

Open http://localhost:3000.

For local development setup, optional Redis/PostgreSQL profiles, and Windows instructions, see the Getting Started guide.

AI Providers

Provider Environment Variable Default Model
Anthropic Claude ANTHROPIC_API_KEY claude-sonnet-4-20250514
OpenAI OPENAI_API_KEY gpt-4o-mini
Google Gemini GOOGLE_API_KEY gemini-2.5-flash
OpenRouter OPENROUTER_API_KEY openrouter/auto
Ollama (local, free) AI_PROVIDER=local mistral:7b

Auto-detects in order: Anthropic → OpenAI → Google → OpenRouter → Ollama. Switch at any time from the header dropdown or Settings page.

Full setup guide including Ollama: AI Providers →

Production deployments (Render / Fly / Railway)

⚠️ Important: free-tier root filesystems are usually ephemeral. If your SQLite DB lives on ephemeral storage, every redeploy can wipe accounts, projects, tests, and runs.

Use the included render.yaml Blueprint on Render to mount a 1 GB Persistent Disk at /app/backend/data and set DB_PATH=/app/backend/data/sentri.db.

If you plan to run multiple instances, prefer managed Postgres and set DATABASE_URL=postgres://... instead of SQLite.

Documentation
Getting Started Installation, first steps, optional services
Architecture Pipeline, data flow, design decisions
Auto-Approval Confidence-based auto-approval, threshold tuning, audit trail, kill-switch
Self-Healing Selector waterfall, healing history, failure classification
Test Dials Strategy, workflow, quality, format, language options
API Reference Full REST API with request/response examples
Production Checklist Security, infrastructure, and deployment hardening
Environment Variables Complete backend and frontend variable reference
Manual QA Guide End-to-end manual test plan, Golden E2E happy path, per-feature checks

Contributing

Contributions are welcome. Please read CONTRIBUTING.md before opening a pull request.

Before you start:

  • Check open issues and ROADMAP.md to avoid duplicating in-progress work
  • For significant changes, open an issue first to discuss the approach

Workflow:

  1. Fork the repository and create a branch: feature/<description> or fix/<description>
  2. Read AGENT.md — it covers architecture, conventions, and what not to do
  3. Read STANDARDS.md when writing new code
  4. Run the test suite before submitting: cd backend && npm test and cd frontend && npm run build
    • For user-visible changes, also walk the affected sections of QA.md — at minimum the Golden E2E Happy Path
  5. Follow Conventional Commits for commit and PR title format — the release pipeline uses this to determine version bumps automatically
  6. Update docs/changelog.md under ## [Unreleased] for any user-visible change
  7. Read REVIEW.md before opening the PR

Code quality: every PR that adds or modifies backend logic must include tests. PRs without adequate coverage will not be merged. See REVIEW.md for the full requirements table.

License

MIT — see LICENSE for details.