@alltoken/ai

Official TypeScript SDK for AllToken — one API for OpenAI, Anthropic, and 100+ models.

npm install @alltoken/ai

Requires Node.js 18+ (for built-in fetch). Works in browsers, Deno, Bun, Cloudflare Workers.

Quick start

import { AllToken } from '@alltoken/ai';

const client = new AllToken({ apiKey: process.env.ALLTOKEN_API_KEY! });

// OpenAI-compatible surface (maps to /v1)
const { data } = await client.openai.raw.POST('/chat/completions', {
  body: {
    model: 'gpt-4o',
    messages: [{ role: 'user', content: 'Hello!' }],
  },
});

// Anthropic-compatible surface (maps to /anthropic)
const { data: data2 } = await client.anthropic.raw.POST('/messages', {
  body: {
    model: 'claude-sonnet-4',
    max_tokens: 1024,
    messages: [{ role: 'user', content: 'Hello!' }],
  },
});

The same API key works for both surfaces. Model catalog: alltoken.ai/models.

Configuration

new AllToken({
  apiKey: '...',                        // required
  baseURL: 'https://api.alltoken.ai',   // optional, defaults to production
  fetch: customFetch,                   // optional, defaults to globalThis.fetch
  defaultHeaders: { 'X-My-Tag': 'a' },  // optional, merged into every request
});

API surface

| Field | Spec | Base URL | |---|---|---| | client.openai.raw | chat.yml (OpenAI-compatible) | https://api.alltoken.ai/v1 | | client.anthropic.raw | anthropic.yml | https://api.alltoken.ai/anthropic |

.raw is the typed openapi-fetch client. All routes, request bodies, and response types are auto-generated from the OpenAPI specs — you get full IDE autocomplete with no manual type wrangling.

Status

v0.1.0 — Scaffold. Types are generated from the spec, the wrapper surface is minimal. Expect breaking changes in 0.x. Production-ready ergonomic helpers (chat.completions.create(...), streaming iterators, retries, etc.) are coming in 0.2.x.

Contributing / Local development

# Clone megaopenrouter as a sibling (for the OpenAPI specs)
git clone [email protected]:ai-innovation-lab/megaopenrouter.git ../megaopenrouter

# Install + regenerate types from specs
npm install
npm run generate

# Build + typecheck
npm run build
npm run typecheck
npm test

Types live in src/generated/{chat,anthropic}.d.ts — these are committed so users who install from npm don't need to run codegen.

License

MIT