OpenAPI generator

This package generates easy-to-use and type-safe frontend API clients from OpenAPI specifications.

Features:

• TypeScript schema generator — generates full TypeScript types from OpenAPI 3 / Swagger 2 specs
• Fetch client — type-safe fetch-based HTTP client with interceptors
• SSE streaming — first-class Server-Sent Events support for AI/streaming APIs
• React hooks — useStreamChat for AI chat, useRequest for data fetching
• Miniapp client — WeChat miniapp client (wx.request)

Installation

npm i @doremijs/o2t
# pnpm i @doremijs/o2t
# yarn add @doremijs/o2t
# bun i @doremijs/o2t

Usage — Code Generation

1. Run npx o2t init to create a o2t.config.mjs configuration file, or create it manually:

import { defineConfig } from '@doremijs/o2t'
export default defineConfig({
  specUrl: 'https://petstore.swagger.io/v2/swagger.json'
})

2. Run the generator:

npx o2t generate typescript

3. The generated code will be in the src/api directory.

4. Create src/api/index.ts and set up the client:

import { createFetchClient } from '@doremijs/o2t/client'
import type { OpenAPIs } from './schema'

export const client = createFetchClient<OpenAPIs>({
  requestInterceptor(request) {
    const token = localStorage.getItem('access_token')
    if (!request.url.startsWith('/api/auth') && token) {
      request.init.headers.Authorization = `Bearer ${token}`
    }
    return request
  },
  responseInterceptor(request, response) {
    return response
  },
  errorHandler(request, response, error) {
    console.error(request, response, error)
  }
})

If you get an import error, set compilerOptions.moduleResolution to bundler and compilerOptions.module to ESNext in your tsconfig.json.

5. Make type-safe requests:

const result = await client.get('/pet/{petId}', {
  params: { petId: 1 },
  query: { status: 'available' }
})
if (!result.error) {
  console.log(result.data)  // Fully typed!
}

SSE Streaming

The package includes a production-grade SSE (Server-Sent Events) streaming engine, designed for AI chat APIs and real-time data.

Quick Start

import { createFetchClient } from '@doremijs/o2t/client'
import { processStream } from '@doremijs/o2t/client/stream'

const client = createFetchClient()

// 1. Make a request (streaming endpoint)
const result = await client.post('/api/chat', {
  body: { stream: true, model: 'gpt-4', messages: [...] }
})

// 2. Process the SSE stream
if (!result.error && result.response) {
  await processStream(result.response, {
    onData: (chunk) => console.log('Received:', chunk),
    onComplete: () => console.log('Stream complete')
  })
}

Three Consumption Styles

API	Style	When to use
processStream(response, callbacks)	Callback	Simplest — just need onData/onError/onComplete
iterateStream(response)	Async iterator	Need fine-grained control with for await...of
accumulateStream(response)	Collect all	Short streams where you want all results at once

Streaming with React

import { useStreamChat } from '@doremijs/o2t/client/react'

function Chat() {
  const { messages, isLoading, send, abort } = useStreamChat({
    service: async (params, signal) => {
      const res = await fetch('/api/chat', {
        method: 'POST',
        body: JSON.stringify(params),
        signal
      })
      return res
    },
    localTransform: (params) => ({
      role: 'user',
      content: params.message
    }),
    streamTransform: ({ chunks }) => ({
      role: 'assistant',
      content: chunks.map(c => c?.choices?.[0]?.delta?.content || '').join('')
    })
  })

  return (
    <div>
      {messages.map(msg => <div key={msg.id}>{msg.data.content}</div>)}
      <button onClick={() => send({ message: 'Hello' })}>Send</button>
      {isLoading && <button onClick={abort}>Stop</button>}
    </div>
  )
}

Configuration Options

The stream engine handles all common SSE variants out of the box, but you can customize:

// Skip non-JSON events (like [DONE] markers silently)
for await (const event of iterateStream(response, {
  nonJSONBehavior: 'skip',
})) { }

// Custom termination markers
await processStream(response, callbacks, {
  terminationMarkers: ['[END]', '[DONE]']
})

// Disable termination detection
const chunks = await accumulateStream(response, {
  terminationMarkers: false
})

SSE Compatibility

The parser is fully compliant with the W3C SSE specification and tested against:

• ✅ Standard SSE (data: ...\n\n)
• ✅ CRLF line endings (\r\n\r\n)
• ✅ Multiple data: lines per event
• ✅ event: / id: / retry: fields
• ✅ Comment lines (: ...)
• ✅ UTF-8 BOM
• ✅ OpenAI [DONE] termination marker
• ✅ Non-JSON data handling
• ✅ Error events (event: error)
• ✅ Cross-chunk boundaries (streaming split across multiple reads)

Configuration

The defineConfig function accepts:

Option	Type	Default	Description
specUrl	string	—	OpenAPI specification URL
isVersion2	boolean	auto	Force Swagger 2.0 mode
outputDir	string	"src/api"	Output directory for generated code
tempFilePath	string	"node_modules/.o2t/openapi.json"	Temporary file for downloaded spec
preferUnknownType	'any' | 'unknown'	"any"	Type for unknown schemas
customHeaders	Record<string, string>	—	Custom headers for spec download
basicAuth	{ username, password }	—	Basic auth credentials

Development

bun install

Create o2t.config.mjs:

import { defineConfig } from './src'
export default defineConfig({
  specUrl: 'https://generator.swagger.io/api/swagger.json'
})

Run in development mode:

bun dev generate typescript

Testing

bun test                          # All tests
bun test:stream                   # SSE stream tests
bun test:fetch                    # Fetch client tests
bun test:react                    # React hooks tests
bun test:petstore                 # PetStore integration tests

Example App

An example React + Vite app is available in the example/ directory:

cd example
bun install
bun dev

Includes:

• AI chat with streaming responses (ChatPage)
• PetStore API demo (PetStorePage)
• Full type-safe integration

CI/CD Publishing

This repository includes a GitHub Actions workflow at .github/workflows/npm-publish.yml.

It will publish the package to npm when all of the following are true:

• code is pushed to the main branch
• package.json or .github/workflows/npm-publish.yml changed in that push
• the version field in package.json is different from the version currently published on npm
• CI tests and build pass

The workflow uses npm trusted publishing with OIDC and runs:

npm publish --provenance --access public

To keep release publishing deterministic, CI uses the local test suite from npm run test:ci. The live PetStore integration tests remain available through npm run test:petstore.

One-time npm setup

GitHub Actions trusted publishing also requires one manual setup step on npm:

1. Open the package page on npm
2. Go to Settings -> Trusted Publisher
3. Choose GitHub Actions
4. Configure:
   • owner: doremijs
   • repository: openapi-generator
   • workflow file: npm-publish.yml
   • branch: main

After that, as long as the version in package.json is newer than the version currently published on npm, pushes to main that touch package.json or the workflow file can trigger an authenticated publish without storing an npm access token in GitHub secrets.

License

LGPL-3.0-or-later