Ollama is open-source software that makes running open language models on a personal computer as simple as possible. With one command you download and run a model such as Gemma, Qwen, Llama or DeepSeek, then chat with it in the command line or the desktop app. Most importantly, it creates a local OpenAI-compatible API, so programs, plugins and tools written for cloud services can connect to the model on your own machine by changing the address. Its main difference from competitors is its simplicity and its popularity with developers; almost every important AI tool supports it.

What is Ollama good for?#

  • Working with confidential documents: a lawyer or accountant summarizes contracts and financial reports without sending them to an outside server.
  • An offline coding assistant: you connect code-editor extensions and JetBrains IDEs to a local model so you get code suggestions even without internet.
  • Building a product without API costs: a PHP or Python developer adds text summarizing or classification with a local model in a company's internal software.
  • Automation with n8n: you build an agent that sorts emails or tickets using a local model.
  • Learning and experiments: a student tries different models side by side and learns about RAG and agents.

Strengths and weaknesses#

Strengths:

  • Very simple installation and running on Windows, Mac and Linux.
  • A local OpenAI-compatible API and wide support in other tools.
  • Full privacy; once a model is downloaded, no data leaves your machine.
  • A large library of ready-made models.

Weaknesses:

  • Models that run on an ordinary machine are lower in quality than large cloud chatbots.
  • Large models need expensive hardware and run slowly on an ordinary laptop.
  • The main work is on the command line; the desktop app has limited features.

Free tier and pricing#

Ollama itself is open source and completely free, and running locally costs nothing beyond electricity and hardware. It also has an optional cloud service that runs large models on its own servers, with a paid subscription. Check the official site for current pricing.

Language support#

Quality in languages other than English depends entirely on the model: very small models write them badly, while mid-sized and large models from families such as Gemma and Qwen do acceptably. Once a model is downloaded, Ollama runs without internet. Only download model files from sources you trust.

Getting started in three steps#

  1. Install the version for your operating system from the official site.
  2. In the command line, pull and run a small model, for example from the Gemma family.
  3. If the speed and quality are satisfactory, connect it to your code editor or your own program.

Ollama alternatives#

  • LM Studio: the same job with a full graphical interface, for people who aren't comfortable with the command line.
  • Hugging Face: the largest repository of open models, for finding specialized ones.
  • Gemma: one of the best model families to run with Ollama on an ordinary machine.
  • Llama: Meta's family of open models with a large ecosystem.

For a complete guide, read running AI on a personal computer, and for use on a website, setting up an open model on a server for your site.