cy-ai

Note

This package is under development so expect breaking changes in future releases.

Cypress AI

🧪 Cypress AI command that generates E2E tests using an LLM (Large Language Model):

cy.ai(string)

Prerequisites

Install

NPM:

npm install cy-ai --save-dev

Yarn:

yarn add cy-ai --dev

Usage

If you're using TypeScript, import the command using ES2015 syntax:

echo "import 'cy-ai'" >> cypress/support/commands.ts

Or if you're using JavaScript, use CommonJS require:

echo "require('cy-ai')" >> cypress/support/commands.js

Start the Ollama server:

ollama serve

Download the LLM:

ollama pull qwen2.5-coder

Write a test:

// cypress/e2e/example.cy.js
it('visits example.com', () => {
  cy.ai('go to https://example.com and see heading "Example Domain"')
})

Tip

If you're running Chrome, disable chromeWebSecurity so the LLM requests aren't blocked by CORS:

// cypress.config.js
import { defineConfig } from 'cypress'

export default defineConfig({
  chromeWebSecurity: false,
})

cy.ai

Generate Cypress tests with AI:

cy.ai(string[, options])

llm

LangChain Runnable to invoke. Defaults to a prompt template using the Ollama model qwen2.5-coder.

Use a different large language model:

import { Ollama } from '@langchain/ollama'
import { prompt } from 'cy-ai'

const llm = new Ollama({
  model: 'codellama',
  numCtx: 16384,
})

const chain = prompt.pipe(llm)

cy.ai('prompt', { llm: chain })

Or customize the template:

import { PromptTemplate } from '@langchain/core/prompts'
import { Ollama } from '@langchain/ollama'

const llm = new Ollama({
  model: 'codellama',
  numCtx: 16384,
})

const prompt = PromptTemplate.fromTemplate(`
You are writing an E2E test step with Cypress.

Rules:
1. Return JavaScript Cypress code without "describe" and "it".

Task: {task}

HTML:
\`\`\`html
{html}
\`\`\`
`)

const chain = prompt.pipe(llm)

cy.ai('prompt', { llm: chain })

Important

Don't forget to pull the Ollama model:

ollama pull codellama

log

Whether to display the command logs. Defaults to true:

cy.ai('prompt', { log: true })

Hide Cypress and console logs:

cy.ai('prompt', { log: false })

regenerate

Whether to regenerate the Cypress step with AI. Defaults to false:

cy.ai('prompt', { regenerate: false })

Regenerate the Cypress step with AI:

cy.ai('prompt', { regenerate: true })

timeout

Time to wait in milliseconds. Defaults to 2 minutes:

cy.ai('prompt', { timeout: 120000 })

Set timeout to 5 minutes:

cy.ai('prompt', { timeout: 1000 * 60 * 5 })

cy.aiConfig

Configure global options for cy.ai:

cy.aiConfig(options)

options

Override default options:

cy.aiConfig({
  llm: chain,
  log: false,
  regenerate: true,
  timeout: 1000 * 60 * 3, // 3 minutes
})

Set timeout to 5 minutes:

cy.aiConfig({
  timeout: 1000 * 60 * 5,
})

How It Works

A prompt is created from your task, the HTML body, and the template.
The prompt is sent to the LLM server.
The LLM server responds with Cypress code.
The Cypress code is cleaned and run.
If the steps pass, the code is saved to cypress/e2e/**/__generated__/*.json.
If the steps fail, an error is thrown and the LLM response can be inspected in the browser Console.

When running tests, if the generated Cypress code exists, the command will reuse the existing code.

To regenerate a step, enable the regenerate option or delete the generated code in cypress/e2e/**/__generated__/*.json.

Warning

If you have tests with duplicate or identical titles (describe and it), it could cause the generated tests to fail.

Release

Release is automated with Release Please.

Resources

cypress-ai-demo

License

MIT