OPEN SOURCE · MIT
TestivAI Documentation
Visual regression testing that runs locally.
No account, no API key. An AI model is optional, and it is yours.
npm install -D @testivai/witness-playwright @playwright/test → npx playwright testGetting Started
Install the adapter, add one witness() call, run the tests, open the report.
How It Works
Pixel, DOM and computed-style comparison with element attribution, and what the noise hint means.
AI Agents (MCP)
Connect Claude Code, Cursor or any MCP client. The model explains each diff; in Claude Code you approve from the conversation, with a confirmation prompt.
Claude Code plugin
The whole report inside Claude Code, screenshots included, with a pane of each snapshot's verdict. Installed from a terminal; cloud sessions cannot load plugins.
GitHub Action
The diff as a PR comment and check, /testivai approve from the pull request, each run's report on GitHub Pages linked from the comment, and Claude's triage of each diff.
CLI Reference
testivai init, witness, report (--json, exit codes), approve, run and env.
Supported Frameworks
Setup and examples per framework.