cocoon-ai-provider

AI SDK v6 provider for Cocoon.

Install

pnpm add cocoon-ai-provider ai

Quick Start

import { cocoon } from 'cocoon-ai-provider';
import { generateText } from 'ai';

const result = await generateText({
  model: cocoon('Qwen/Qwen3-32B'),
  prompt: 'Write a one-line welcome message.',
});

console.log(result.text);
await cocoon.close();

Streaming

import { cocoon } from 'cocoon-ai-provider';
import { streamText } from 'ai';

const result = streamText({
  model: cocoon('Qwen/Qwen3-32B'),
  prompt: 'Write a short product tagline.',
});

for await (const textPart of result.textStream) {
  process.stdout.write(textPart);
}

await cocoon.close();

Runnable Example

This repository includes a runnable end-to-end example at examples/basic.mjs.

COCOON_WALLET="your 24-word mnemonic" pnpm example:basic

Optional:

• COCOON_MODEL_ID to override the model (default: Qwen/Qwen3-32B)
• Any other Cocoon env vars described below

Custom Provider Instance

You can pass an existing Cocoon client or let the provider build one from clientOptions.

import { createCocoon } from 'cocoon-ai-provider';

const provider = createCocoon({
  clientOptions: {
    wallet: process.env.MNEMONIC!,
    secretString: process.env.SECRET,
    proxyUrl: process.env.PROXY_URL,
  },
  defaults: {
    maxCoefficient: 1.2,
    timeout: 30_000,
  },
});

Provider Options

Use call-level Cocoon options with providerOptions.cocoon:

import { generateText } from 'ai';
import { cocoon } from 'cocoon-ai-provider';

const result = await generateText({
  model: cocoon('Qwen/Qwen3-32B'),
  prompt: 'Summarize this in one sentence.',
  providerOptions: {
    cocoon: {
      maxCoefficient: 1.1,
      timeout: 20_000,
    },
  },
});

Call-level providerOptions.cocoon override provider-level defaults.

Environment Variables

When createCocoon() is called without client and without clientOptions, the provider builds a Cocoon client from env values.

• wallet: COCOON_WALLET or MNEMONIC (required)
• secretString: COCOON_SECRET_STRING or SECRET
• proxyUrl: COCOON_PROXY_URL or PROXY_URL
• tonEndpoint: COCOON_TON_ENDPOINT or TON_ENDPOINT
• tonV4Endpoint: COCOON_TON_V4_ENDPOINT or TON_V4_ENDPOINT
• network: COCOON_NETWORK (mainnet or testnet)
• timeout: COCOON_TIMEOUT (positive number)
• useTls: COCOON_USE_TLS (true/false or 1/0)
• autoRegisterOnLongAuth: COCOON_AUTO_REGISTER_ON_LONG_AUTH (true/false or 1/0, default false)
• longAuthRegisterAmountTon: COCOON_LONG_AUTH_REGISTER_AMOUNT_TON
• TLS credential values: COCOON_TLS_CERT + COCOON_TLS_KEY
• TLS credential files: COCOON_TLS_CERT_PATH + COCOON_TLS_KEY_PATH

Runtime and Limitations

• Node.js >=18 only.
• v1 supports language-model generation and streaming only.
• Tool calling is not supported.
• Multimodal file input is not supported.
• Embedding and image model methods intentionally throw NoSuchModelError.
• Abort behavior is best-effort. Cocoon cancellation is limited by the underlying SDK.

Testing

pnpm test

Optional real-network integration tests:

pnpm test:integration