How to create a simple DeekSeek NodeJS server using a local model

This is a quick 5 minute overview of how to setup a nodeJS server that responds to prompts using a local Deepseek model, or any model supported by Ollama.

This is based off of the instructions found here.

Download and install Ollama

Pull a Deepseek model of your choice. You can find more models here
ollama pull deepseek-r1:7b

Initialise your project
pnpm init

Install Vercel AI
pnpm install ai

Install the ollama provider
pnpm install ollama-ai-provider

Install fastify
pnpm install fastify

Install Zod
pnpm install zod

Create a index.ts file and paste in the following code

import { generateText } from 'ai';
import { createOllama } from 'ollama-ai-provider';
import createFastify from 'fastify';
import z from 'zod'

const fastify = createFastify();

const promptSchema = z.object({
  prompt: z.string()
})

const ollama = createOllama({
  baseURL: 'http://localhost:11434/api',
});

fastify.post('/prompt', async (request, reply) => {
  const promptResult = promptSchema.safeParse(request.body);

  if (promptResult.error) {
    console.log(promptResult.error)
    return reply
      .code(500)
      .send();
  }

  const result = await generateText({
    model: ollama('deepseek-r1:7b'),
    prompt: promptResult.data.prompt
  });

  return { answer: result.text }
})

await fastify.listen({ port: 3000 }, () => {
  console.log('listening on port 3000')
})

Make sure you have tsx installed and run
npx tsx index.ts

You can curl the endpoint to get a response
>curl -X POST http://localhost:3000/prompt -H "Content-Type: application/json" --json "{\"prompt\": \"Tell me a story\"}"