Ollama Adapter

The Ollama adapter connects to a local or remote Ollama instance. It uses the ollama npm package's /browser sub-export — this strips Node-only node:fs code that would cause bundler errors in Vite, Next.js, or any browser-targeted build.

Installation

Code

npm install tekimax-ts

Usage

Code

import { Tekimax, OllamaProvider } from 'tekimax-ts';

const client = new Tekimax({
  provider: new OllamaProvider({
    // Defaults to 127.0.0.1 (not "localhost") to avoid IPv6 DNS
    // resolution issues on macOS and some Linux distros.
    host: 'http://127.0.0.1:11434',
  })
});

const result = await client.text.chat.completions.create({
  // Model must be pulled first: `ollama pull llama3`
  model: 'llama3',
  messages: [{ role: 'user', content: 'Why is the sky blue?' }]
});

console.log(result.message.content);

Authentication & Cloud Support

You can connect to a remote Ollama instance (e.g., behind a reverse proxy or Ollama Cloud) by providing an apiKey. The adapter injects an Authorization: Bearer <key> header via a custom fetch wrapper — the Ollama JS SDK doesn't natively support auth tokens.

Code

const client = new Tekimax({
  provider: new OllamaProvider({
    host: 'https://your-ollama-instance.com',
    apiKey: 'ollama_key_...'
  })
});

Streaming

Local models support streaming out of the box with the standard interface.

Code

const stream = client.text.chat.completions.createStream({
  model: 'mistral',
  messages: [{ role: 'user', content: 'Count to 10 efficiently.' }]
});

for await (const chunk of stream) {
  process.stdout.write(chunk.delta);
}

Reasoning (Thinking)

Ollama has native support for the thinking field on compatible models. Set think: true to capture the reasoning trace.

Code

const result = await client.text.chat.completions.create({
  model: 'deepseek-r1',
  messages: [{ role: 'user', content: 'Solve: if 2x + 3 = 11, what is x?' }],
  think: true // Ollama passes this directly to the model
});

console.log(result.message.thinking); // "2x + 3 = 11, 2x = 8, x = 4"
console.log(result.message.content);  // "x = 4"

Ollama supports vision with models like llava. Pass images as base64 data URIs in the message content.

Code

const result = await client.text.chat.completions.create({
  model: 'llava',
  messages: [{
    role: 'user',
    content: [
      { type: 'text', text: 'What is in this image?' },
      { type: 'image_url', image_url: { url: 'data:image/png;base64,iVBOR...' } }
    ]
  }]
});

console.log(result.message.content);

Tool Calling

Tool calling is supported on compatible Ollama models (e.g., llama3, mistral).

Code

const result = await client.text.chat.completions.create({
  model: 'llama3',
  messages: [{ role: 'user', content: 'What is the weather in Berlin?' }],
  tools: [{
    type: 'function',
    function: {
      name: 'get_weather',
      description: 'Get current weather',
      parameters: { type: 'object', properties: { location: { type: 'string' } } }
    }
  }]
});

// Ollama doesn't provide unique tool call IDs, so the adapter generates them.
console.log(result.message.toolCalls);

Notes

Why ollama/browser? The default ollama import pulls in node:fs for file-based model management. This breaks client-side bundlers (Vite, webpack, Next.js). The /browser sub-export strips that code while keeping the chat API intact.
Why 127.0.0.1? Using the IP literal instead of localhost avoids DNS resolution delays and failures on systems where localhost resolves to ::1 (IPv6) but Ollama only binds to 0.0.0.0 (IPv4).
Model Availability: Models must be pulled locally before use (ollama pull <model>). The SDK does not auto-pull models.

Ollama Adapter

Installation

Usage

Authentication & Cloud Support

Streaming

Reasoning (Thinking)

Vision (Multi-Modal)

Tool Calling

Notes

On this page