Ollama Plugin

The genkitx-ollama plugin enables you to run AI models locally using Ollama. This is ideal for development, testing, and applications that need to run models on-premise without external API calls.

Installation

npm install genkitx-ollama

Prerequisites

Install Ollama: Download and install from ollama.ai
Pull models: Download the models you want to use

# Pull a model (example: Gemma)
ollama pull gemma

# Pull other popular models
ollama pull llama3
ollama pull mistral
ollama pull codellama

Start Ollama server: The server runs automatically after installation, or start it manually:

ollama serve

Default server address: http://localhost:11434

Basic Setup

import { genkit } from 'genkit';
import { ollama } from 'genkitx-ollama';

const ai = genkit({
  plugins: [
    ollama({
      models: [{ name: 'gemma' }],
      serverAddress: 'http://127.0.0.1:11434', // default
    }),
  ],
});

const { text } = await ai.generate({
  prompt: 'Tell me about local AI models',
  model: 'ollama/gemma',
});

console.log(text);

Configuration

Plugin Options

ollama({
  models: [
    { 
      name: 'gemma',
      type: 'chat',           // 'chat' or 'generate' (default: 'chat')
      supports: {
        tools: true,          // Enable tool calling
      },
    },
    { 
      name: 'llama3',
      type: 'chat',
    },
    {
      name: 'codellama',
      type: 'generate',       // Use generate API for non-chat models
    },
  ],
  embedders: [
    {
      name: 'nomic-embed-text',
      dimensions: 768,        // Required for embedders
    },
  ],
  serverAddress: 'http://localhost:11434',
  requestHeaders: {         // Optional custom headers
    'Authorization': 'Bearer token',
  },
})

Model Configuration

const response = await ai.generate({
  model: 'ollama/gemma',
  prompt: 'Your prompt',
  config: {
    temperature: 0.8,        // Default: 0.8 (0.0-1.0)
    topK: 40,                // Default: 40
    topP: 0.9,               // Default: 0.9 (0.0-1.0)
    maxOutputTokens: 2048,   // Maps to num_predict
    stopSequences: ['END'],  // Stop generation sequences
  },
});

Popular Models

Chat Models

ollama({
  models: [
    { name: 'llama3' },        // Meta's Llama 3
    { name: 'gemma' },         // Google's Gemma
    { name: 'mistral' },       // Mistral AI
    { name: 'mixtral' },       // Mistral's mixture-of-experts
    { name: 'phi3' },          // Microsoft's Phi-3
    { name: 'qwen2' },         // Alibaba's Qwen
  ],
})

Code Models

ollama({
  models: [
    { name: 'codellama', type: 'chat' },
    { name: 'deepseek-coder' },
    { name: 'starcoder2' },
  ],
})

Embedding Models

ollama({
  embedders: [
    { name: 'nomic-embed-text', dimensions: 768 },
    { name: 'mxbai-embed-large', dimensions: 1024 },
    { name: 'all-minilm', dimensions: 384 },
  ],
})

Usage Examples

Text Generation

import { genkit } from 'genkit';
import { ollama } from 'genkitx-ollama';

const ai = genkit({
  plugins: [
    ollama({
      models: [{ name: 'llama3' }],
    }),
  ],
});

const response = await ai.generate({
  model: 'ollama/llama3',
  prompt: 'Explain how local AI models work',
});

console.log(response.text);

Multi-turn Conversation

const response = await ai.generate({
  model: 'ollama/gemma',
  messages: [
    { role: 'user', content: [{ text: 'What is Ollama?' }] },
    { role: 'model', content: [{ text: 'Ollama is a tool for running AI models locally.' }] },
    { role: 'user', content: [{ text: 'How do I install it?' }] },
  ],
});

console.log(response.text);

Tool Calling

import { z } from 'genkit';

const getWeather = ai.defineTool(
  {
    name: 'getWeather',
    description: 'Get current weather for a location',
    inputSchema: z.object({
      location: z.string(),
    }),
    outputSchema: z.string(),
  },
  async ({ location }) => {
    return `Weather in ${location}: Sunny, 72°F`;
  }
);

const response = await ai.generate({
  model: 'ollama/llama3',
  prompt: 'What\'s the weather in San Francisco?',
  tools: [getWeather],
});

console.log(response.text);

Image Input (Multimodal)

const response = await ai.generate({
  model: 'ollama/llava',  // Use a multimodal model
  prompt: [
    { text: 'What do you see in this image?' },
    { media: { url: 'data:image/jpeg;base64,...' } },  // Base64 image
  ],
});

console.log(response.text);

Embeddings

import { genkit } from 'genkit';
import { ollama } from 'genkitx-ollama';

const ai = genkit({
  plugins: [
    ollama({
      embedders: [
        { name: 'nomic-embed-text', dimensions: 768 },
      ],
    }),
  ],
});

const embeddings = await ai.embed({
  embedder: ollama.embedder('nomic-embed-text'),
  content: 'Text to embed for semantic search',
});

console.log(embeddings[0].embedding); // Array of 768 numbers

Using in Flows

import { z } from 'genkit';

const codeReviewFlow = ai.defineFlow(
  {
    name: 'codeReview',
    inputSchema: z.object({
      code: z.string(),
      language: z.string(),
    }),
    outputSchema: z.string(),
  },
  async ({ code, language }) => {
    const response = await ai.generate({
      model: 'ollama/codellama',
      prompt: `Review this ${language} code and suggest improvements:\n\n${code}`,
    });
    return response.text;
  }
);

const review = await codeReviewFlow({
  code: 'function add(a, b) { return a + b; }',
  language: 'JavaScript',
});

Direct Model Usage

import { ollama } from 'genkitx-ollama';

// Create model reference
const model = ollama.model('llama3');

// Use directly without Genkit instance
const response = await model({
  messages: [
    {
      role: 'user',
      content: [{ text: 'Hello!' }],
    },
  ],
});

console.log(response);

Advanced Configuration

Custom Server Address

ollama({
  models: [{ name: 'gemma' }],
  serverAddress: 'http://192.168.1.100:11434',  // Remote Ollama server
})

Custom Request Headers

ollama({
  models: [{ name: 'gemma' }],
  requestHeaders: {
    'Authorization': 'Bearer my-token',
    'X-Custom-Header': 'value',
  },
})

// Or use a function for dynamic headers
ollama({
  models: [{ name: 'gemma' }],
  requestHeaders: async (context, input) => {
    return {
      'Authorization': `Bearer ${await getToken()}`,
    };
  },
})

Model-specific Settings

ollama({
  models: [
    {
      name: 'llama3',
      type: 'chat',
      supports: {
        tools: true,           // Enable tool calling
        multiturn: true,       // Multi-turn conversations
        systemRole: true,      // System messages
      },
    },
  ],
})

Model Management

List Available Models

ollama list

Pull New Models

ollama pull llama3
ollama pull gemma:7b        # Specific version
ollama pull codellama:13b   # Larger variant

Remove Models

ollama rm gemma

Show Model Info

ollama show llama3

Best Practices

Choose Appropriate Model Size

7B models - Fast, good for most tasks, 8GB RAM
13B models - Better quality, 16GB RAM recommended
70B+ models - Highest quality, requires 32GB+ RAM

Optimize Performance

// Use smaller context windows
const response = await ai.generate({
  model: 'ollama/gemma',
  prompt: 'Your prompt',
  config: {
    maxOutputTokens: 512,  // Limit response length
  },
});

Handle Errors

try {
  const response = await ai.generate({
    model: 'ollama/gemma',
    prompt: 'Your prompt',
  });
  console.log(response.text);
} catch (error) {
  if (error.message?.includes('ECONNREFUSED')) {
    console.error('Ollama server is not running. Start it with: ollama serve');
  } else {
    console.error('Error:', error);
  }
}

Pre-load Models

Pre-load models to reduce first-request latency:

# Keep model loaded in memory
ollama run gemma
# Press Ctrl+D to exit but keep model loaded

Limitations

Tool calling: Only available on chat API, not generate
Input schema: Tools must have object input schemas
Performance: Depends on local hardware
Model size: Larger models require more RAM and are slower

Troubleshooting

Server Not Running

Error: ECONNREFUSED Solution: Start the Ollama server:

ollama serve

Model Not Found

Error: Model not available Solution: Pull the model first:

ollama pull gemma

Out of Memory

Solution: Use a smaller model or increase system RAM

Slow Performance

Solutions:

Use smaller models (7B instead of 13B)
Reduce maxOutputTokens
Use GPU acceleration if available
Close other applications

Official Plugins

Plugin Development

Documentation Index

​Ollama Plugin

​Installation

​Prerequisites

​Basic Setup

​Configuration

​Plugin Options

​Model Configuration

​Popular Models

​Chat Models

​Code Models

​Embedding Models

​Usage Examples

​Text Generation

​Multi-turn Conversation

​Tool Calling

​Image Input (Multimodal)

​Embeddings

​Using in Flows

​Direct Model Usage

​Advanced Configuration

​Custom Server Address

​Custom Request Headers

​Model-specific Settings

​Model Management

​List Available Models

​Pull New Models

​Remove Models

​Show Model Info

​Best Practices

​Choose Appropriate Model Size

​Optimize Performance

​Handle Errors

​Pre-load Models

​Limitations

​Troubleshooting

​Server Not Running

​Model Not Found

​Out of Memory

​Slow Performance

​Links

Build docs developers (and LLMs) love