Glossary
What is Ollama?
In short
Ollama is a free, open-source program that runs language models on your own computer or server.
Models and products · Ollama
MMOOZZGG.com editorsChecked: 1 min read
For a child
Imagine a record player that can play any record. The record is at your home, and the music plays even when the internet is off. There is a program that works like this player, but instead of records it has smart helpers that answer questions. You download a helper once, and it works right on your computer.
In plain words
Ollama is a program that downloads an open language model and runs it on your own computer. You can then chat with the model in a terminal or connect it to other apps, as if it were a cloud service. The program runs on macOS, Windows and Linux, and there is also an image for Docker. You choose models from the catalog on the project website. For example, an employee summarizes an internal report with a model that runs on a laptop and sends no text to the internet.
For business
The code is under the MIT license, so companies install the program for free on work machines and servers, including a closed network. Requests to a local model need no key and run on your own hardware. This makes it simpler to work with personal data under 152-FZ, Russia’s personal data law. Ollama also offers cloud models with an API key: there the text goes to the company’s servers, and the May 2026 terms list no countries. Hardware sets the cost: model size and answer speed depend on video memory. When many employees use a model at once, choose the serving program by load tests, not by how easy it is to install.
For a specialist
Ollama starts a local REST API on port 11434: its own methods under /api, an OpenAI-compatible interface under /v1 and a /v1/messages path for the Anthropic client. So an app written for a cloud API switches to a local model by changing the base address. Among the supported engines (backends), the README names llama.cpp, which implements integer weight quantization from 1.5 to 8 bits. This way the model takes less memory and runs on NVIDIA graphics cards through CUDA and on Apple Silicon through Metal. Model parameters and the system prompt are set in a Modelfile. Cloud models are called through the same paths on ollama.com, with an API key. The project itself is under the MIT license.
Example
An analyst installs the program on a work laptop, downloads a small open model and connects a script that summarizes CRM exports. Client data does not go to the cloud.
Related terms
Questions and answers
Is Ollama free?
Sources
Not sure whether your company needs this?
AI solutions map