@agenite/ollama

Ollama provider for Agenite, enabling seamless integration with local Ollama models.

Features

• 🚀 Support for all Ollama models (Llama 2, CodeLlama, Mistral, etc.)
• 🛠️ Function/tool calling support
• 🖼️ Image input support (with multimodal models)
• 📝 Rich content types (text, images, tool calls)
• 🔄 Streaming responses
• ⚙️ Extensive model parameter controls

Installation

1npm install @agenite/ollama

Make sure you have Ollama installed and running locally.

Usage

Basic Chat

1import { OllamaProvider } from '@agenite/ollama';
2
3const provider = new OllamaProvider({
4  model: 'llama2',
5  host: 'http://localhost:11434',
6  temperature: 0.7,
7});
8
9const messages = [
10  {
11    role: 'user',
12    content: [{ type: 'text', text: 'What are the main features of Llama 2?' }],
13  },
14];
15
16const response = await provider.generate({ messages });
17console.log(response.content[0].text);

Tool Usage

1const calculatorTool = {
2  name: 'calculator',
3  description: 'Performs basic arithmetic operations',
4  parameters: {
5    type: 'object',
6    properties: {
7      operation: {
8        type: 'string',
9        enum: ['add', 'subtract', 'multiply', 'divide'],
10      },
11      a: { type: 'number' },
12      b: { type: 'number' },
13    },
14    required: ['operation', 'a', 'b'],
15  },
16};
17
18const response = await provider.generate({
19  messages: [
20    {
21      role: 'user',
22      content: [{ type: 'text', text: 'What is 123 multiplied by 456?' }],
23    },
24  ],
25  tools: [calculatorTool],
26});

Image Input

1const provider = new OllamaProvider({
2  model: 'llava',
3  host: 'http://localhost:11434',
4});
5
6const response = await provider.generate({
7  messages: [
8    {
9      role: 'user',
10      content: [
11        {
12          type: 'image',
13          source: {
14            type: 'base64',
15            media_type: 'image/jpeg',
16            data: imageBase64,
17          },
18        },
19        { type: 'text', text: 'What do you see in this image?' },
20      ],
21    },
22  ],
23});

Configuration

The provider accepts the following configuration options:

1interface OllamaConfig {
2  model: string;           // Model name (e.g., 'llama2', 'codellama')
3  host?: string;          // Ollama server URL (default: http://localhost:11434)
4  temperature?: number;   // Generation temperature (0-1)
5  maxTokens?: number;    // Maximum tokens to generate
6  systemPrompt?: string; // System prompt for all conversations
7  parameters?: {         // Additional model parameters
8    mirostat?: number;
9    num_ctx?: number;
10    num_gpu?: number;
11    // ... and more
12  };
13}

Examples

Check out the examples directory for more detailed examples:

• basic-chat.ts: Simple text conversation
• tool-usage.ts: Calculator tool implementation
• image-input.ts: Image analysis with multimodal models

Development

1# Install dependencies
2npm install
3
4# Build
5npm run build
6
7# Run tests
8npm test
9
10# Run examples
11npx ts-node examples/basic-chat.ts

License

MIT