AI-wright: AI steps in Playwright scripts: open-source, Vision-enabled, BYOL (testchimp.io)

ai-wright is an open-source library that injects AI-powered actions and verifications directly into existing Playwright tests instead of replacing them. You keep deterministic, fast, inspectable Playwright scripts and drop in targeted AI steps (e.g., await ai.act(...) or await ai.verify(...)) only where UI behavior is ambiguous or dynamic. That means teams avoid the slowness, flakiness, and vendor lock‑in of fully agentic testing platforms while gaining AI help for messy flows like content checks, fuzzy element selection, or modal handling. Technically, ai-wright is vision-enabled and LLM-agnostic. It combines Set‑Of‑Marks (SoM) annotated screenshots with structured DOM context so reasoning is grounded in what’s actually on screen (visibility, disabled state, ARIA, shadow DOM, layout), and binds SoM IDs to live DOM nodes for precise one‑shot execution — avoiding fragile selector guessing and giant DOM prompts. Its multi-step reasoning loop plans objectives (e.g., “fill login form”), detects blockers, and runs pre-steps to reduce retries and nondeterminism. You can bring your own LLM (OpenAI, Claude, Gemini, self-hosted) or proxy through TestChimp, preventing vendor lock-in. By adding vision-grounded AI only where needed, ai-wright aims to make tests smarter and more resilient while preserving Playwright’s performance, debuggability, and open-source transparency.