Integrating Ollama with the JavaScript SDK

Let us use the official JavaScript library to program Ollama model queries inside our Node.js backends.

1. Installation

Install the official client SDK package:

# Install the Ollama JS/TS SDK
npm install @ollama/ollama

2. Basic Chat Completions (Non-Streaming)

Import ollama and query a model:

// src/services/ollamaService.ts
import ollama from "@ollama/ollama";

export async function askLocalAI(userPrompt: string) {
  try {
    const response = await ollama.chat({
      model: "qwen2.5",
      messages: [{ role: "user", content: userPrompt }],
      // Set options to customize generation parameters
      options: {
        temperature: 0.2,
      },
    });

    return response.message.content;
  } catch (err: any) {
    console.error("Local model query failed:", err.message);
    return "Local AI service error.";
  }
}

3. Streaming Chat responses

To receive response tokens incrementally, iterate over the async generator returned by setting stream: true:

export async function streamLocalAI(userPrompt: string, onTokenCallback: (token: string) => void) {
  const response = await ollama.chat({
    model: "qwen2.5",
    messages: [{ role: "user", content: userPrompt }],
    stream: true,
  });

  // Loop through token generator chunks
  for await (const part of response) {
    const token = part.message.content;
    onTokenCallback(token);
  }
}