@orka-js/ollama

Ollama adapter for OrkaJS — run local models (Llama 3, Mistral, etc.) with zero API costs and full privacy.

Installation

npm install @orka-js/ollama

Requires Ollama running locally:

# Install and start Ollama
ollama serve
ollama pull llama3.2

Quick Start

import { OllamaAdapter } from '@orka-js/ollama'
import { Orka } from '@orka-js/core'

const llm = new OllamaAdapter({
  model: 'llama3.2',
  // baseUrl defaults to http://localhost:11434
})

const orka = new Orka({ llm })

const response = await orka.chat([
  { role: 'user', content: 'Hello!' }
])
console.log(response.content)

Streaming

const stream = await llm.stream([
  { role: 'user', content: 'Write a poem about the sea' }
])

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

Embeddings

// Requires a model with embedding support (e.g. nomic-embed-text)
const vector = await llm.embed('Hello world')

Configuration

| Option | Type | Default | Description | |--------|------|---------|-------------| | model | string | 'llama3.2' | Local model name | | baseUrl | string | 'http://localhost:11434' | Ollama server URL | | temperature | number | 0.7 | Sampling temperature | | maxTokens | number | — | Max tokens in response |

Popular Models

ollama pull llama3.2        # Meta Llama 3.2
ollama pull mistral         # Mistral 7B
ollama pull codellama       # Code-specialized Llama
ollama pull nomic-embed-text # Embeddings model

API

OllamaAdapter

const adapter = new OllamaAdapter(config)

adapter.chat(messages: ChatMessage[])   // Promise<ChatResponse>
adapter.stream(messages: ChatMessage[]) // AsyncIterable<StreamEvent>
adapter.embed(text: string)             // Promise<number[]>

Implements: LLMAdapter, StreamingLLMAdapter

Cost: $0 — all computation runs locally.

Related Packages

• @orka-js/core — Core types and LLMAdapter interface
• @orka-js/openai — OpenAI adapter
• orkajs — Full bundle