@trovec/embedder-ollama

Ollama embeddings adapter for Trovec. Converts text to vector embeddings using a locally running Ollama server.

Zero runtime dependencies — uses Node.js 18+ built-in fetch. No API key required.

Prerequisites

A running Ollama server with an embedding model pulled. See the Ollama setup guide for Docker-based setup, or install Ollama directly:

ollama pull nomic-embed-text

Installation

npm install @trovec/core @trovec/embedder-ollama

Usage

import { create, addWithText, queryByText } from '@trovec/core';
import { createOllamaEmbedder } from '@trovec/embedder-ollama';

const db = await create({
  embedder: createOllamaEmbedder(),
});
// dimensions are automatically resolved from the embedder (768 for the default model)

await addWithText(db, { id: 'doc1', text: 'The cat sat on the mat' });
await addWithText(db, { id: 'doc2', text: 'Dogs love to play fetch' });

const results = await queryByText(db, { text: 'animals sitting', topK: 5 });

Options

createOllamaEmbedder({
  model?: string;        // default: 'nomic-embed-text'
  baseUrl?: string;      // default: 'http://localhost:11434'
  dimensions?: number;   // auto-resolved for known models; required for custom models
})

All options are optional — the defaults work out of the box with a standard Ollama installation. The returned embedder exposes read-only dimensions and model properties. Trovec uses dimensions to auto-configure itself; model is available for logging and diagnostics.

Models

| Model | Dimensions | Size | Notes | |-------|-----------|------|-------| | nomic-embed-text | 768 | ~274 MB | Default, good quality for general use | | mxbai-embed-large | 1024 | ~670 MB | Higher quality, larger model | | all-minilm | 384 | ~45 MB | Lightweight, fast |

Browse more embedding models at ollama.com/search?c=embedding.

For models not in the list above, pass dimensions explicitly:

createOllamaEmbedder({
  model: 'custom-model',
  dimensions: 512,
})

Remote Server

Use baseUrl to point to an Ollama instance on another machine:

createOllamaEmbedder({
  baseUrl: 'http://192.168.1.100:11434',
})

License

MIT