Ollama (Local)

Install

npm install agentfootprint openai

The openai package is required — agentfootprint’s Ollama adapter uses the OpenAI-compatible API under the hood.

Ollama must be running locally. Install Ollama and pull a model:

ollama pull llama3

Usage

import { Agent, ollama, defineTool } from 'agentfootprint';

const provider = ollama('llama3');

const agent = Agent.create({ provider })
  .system('You are a helpful assistant.')
  .build();

const result = await agent.run('Hello');

Switching from raw Ollama API

Before (raw fetch)

const response = await fetch('http://localhost:11434/api/chat', {
  method: 'POST',
  body: JSON.stringify({
    model: 'llama3.3',
    messages: [{ role: 'user', content: 'Hello' }],
  }),
});
const data = await response.json();
console.log(data.message.content);

After (agentfootprint)

import { Agent, ollama } from 'agentfootprint';

const agent = Agent.create({ provider: ollama('llama3.3') })
  .system('You are helpful.')
  .build();

const result = await agent.run('Hello');
console.log(result.content);
console.log(agent.getNarrative());  // full trace, $0 cost

Custom endpoint

const provider = ollama('llama3', {
  baseUrl: 'http://192.168.1.100:11434',  // remote Ollama instance
});

Development workflow

Use Ollama for local development, switch to Claude for production:

const provider = process.env.NODE_ENV === 'production'
  ? anthropic('claude-sonnet-4-20250514')
  : ollama('llama3');

// Same agent, same tools — just different provider
const agent = Agent.create({ provider })
  .system('You are a support agent.')
  .tool(lookupOrder)
  .build();

Supported models

Any model available in your local Ollama installation. Common choices:

Model	Best for
llama3.3	General purpose
llama4	Latest, multimodal
qwen3	Reasoning, multilingual
deepseek-r1	Deep reasoning
gemma3	Fast, lightweight

Ollama uses the OpenAI-compatible API. For tool use, choose models with reliable function calling support (llama3.1+, qwen2.5, mistral-nemo). Smaller models may produce unreliable tool calls.