ainative-ollama

Drop-in replacement for the ollama npm package that adds AINative cloud fallback when local Ollama is unavailable.

How it works

1. When you call ollama.chat(), the package first tries your local Ollama server (http://localhost:11434).
2. If local Ollama is not running (connection refused / timeout), it seamlessly falls back to AINative cloud.
3. On first cloud fallback, it auto-provisions a free account -- no sign-up required.

Zero overhead when local Ollama is running. Zero config to get started.

Install

npm install ainative-ollama ollama

ollama is a peer dependency -- install it alongside.

Usage

import { AINativeOllama } from 'ainative-ollama';

const ollama = new AINativeOllama();

// Works exactly like the ollama package
const response = await ollama.chat({
  model: 'llama3.3',
  messages: [{ role: 'user', content: 'Why is the sky blue?' }],
});

console.log(response.message.content);

CommonJS

const { create } = require('ainative-ollama');

async function main() {
  const ollama = await create();
  const response = await ollama.chat({
    model: 'llama3.3',
    messages: [{ role: 'user', content: 'Hello!' }],
  });
  console.log(response.message.content);
}

main();

Free cloud models

When falling back to AINative cloud, these models are available for free:

| Ollama model name | Cloud model | Parameters | |---|---|---| | llama3.3, llama3.1, llama3, llama2 | Meta Llama 3.3 70B Instruct | 70B | | qwen, qwen2.5, qwen3, qwen3-coder | Qwen3 Coder Flash | -- | | deepseek, deepseek-coder, deepseek-r1 | DeepSeek 4 Flash | -- | | codellama, starcoder, codegemma | Qwen3 Coder Flash | -- | | mistral, mixtral, gemma, phi | Meta Llama 3.3 70B Instruct | 70B |

Any unrecognized model name is passed through as-is to the AINative API.

Configuration

Environment variables

| Variable | Description | |---|---| | AINATIVE_API_KEY | Skip auto-provisioning, use this API key directly | | AINATIVE_API_URL | Override the AINative API base URL |

Constructor options

const ollama = new AINativeOllama({
  // Standard ollama options
  host: 'http://localhost:11434',

  // AINative-specific options
  ainativeApiKey: 'your-api-key',     // Skip auto-provisioning
  ainativeApiUrl: 'https://...',       // Custom API URL
  cloudOnly: true,                     // Skip local Ollama entirely
});

API

chat(request)

Same signature as ollama.chat(). Returns an Ollama-compatible response object.

generate(request)

Same signature as ollama.generate(). Falls back to cloud chat completions.

list()

Returns merged list of local Ollama models and AINative cloud models. Cloud models have details._source: 'ainative-cloud'.

Auto-provisioning

On first cloud fallback (when no AINATIVE_API_KEY is set), the package:

1. Calls the AINative instant-db endpoint to create a free account
2. Prints the API key and a claim URL to the console
3. Sets AINATIVE_API_KEY in process.env for the rest of the session

Visit the claim URL to convert your temporary account to a permanent one.

License

MIT