Get up and running with large language models, locally.

Run Llama, Mistral, Gemma, and dozens more — on your own laptop, with a single command.

curl -fsSL https://ollama.com/install.sh | sh

Automate your work.

Run any model, locally or in the cloud. Build agents, automate workflows, ship apps — without sending a single token to a third party.

The CLI is the API. Start a model in one line; stream tokens in the next.

ollama run llama3.2

~ — ollama

# run a model locally

$ ollama run llama3.2

pulling manifest...

verifying sha256 digest ✓

writing manifest ✓

success

>>> Why is the sky blue?

Rayleigh scattering: shorter wavelengths

scatter more than longer ones

~ — scale cloud

# same command, datacenter GPUs

$ ollama run llama3.2:70b --cloud

connecting to ollama cloud...

attached gpu: H100 80G ✓

streaming response...

>>> Summarize this 200-page PDF.

Reading 423,991 tokens...

Done in 14.2s

Start local.
Scale cloud.

The same model, the same prompt, the same CLI — running on your laptop or on an H100. Switch with a flag, never with a rewrite.

No vendor lock-in. Models stay portable: pull once, run anywhere.

--cloud

Pricing

Local is free. Cloud is per-token. Pick the tier that fits how you build.

Free

For local development.

$0/ month

Download

Run unlimited models locally
Full CLI and HTTP API
Community support

Pro

Run hosted models when you outgrow local.

$20/ month

Get Pro

Everything in Free, plus:
30M tokens per month
Access cloud-hosted 70B models
Faster pulls and pre-warmed models

Max

Solve harder tasks, faster.

$80/ month

Get Max

Everything in Pro, plus:
200M tokens per month
Access larger models on data-center-grade hardware
Priority routing and dedicated capacity

Frequently asked questions

Is Ollama really free to run locally?

Yes. The Ollama CLI is open source under the MIT license and free for personal and commercial use. You only pay if you opt into the hosted cloud tier.

Which models can I run?

Anything in the Ollama library — Llama 3, Mistral, Gemma, Phi, Qwen, DeepSeek, and many more. You can also bring your own GGUF models with a Modelfile.

Does Ollama see my prompts?

No. When you run locally, everything stays on your machine. When you opt into the cloud tier, your prompts and outputs are encrypted in transit and never used to train models. Contact us at hello@ollama.com for the full data policy.

Can I cancel anytime?

Yes. Cancel from your billing dashboard — your account drops back to the free local tier at the end of the cycle.

How does GPU support work?

Ollama auto-detects Metal on Apple Silicon, CUDA on NVIDIA, and ROCm on AMD. Falls back to CPU otherwise. No driver flags, no env vars.

The reference.

One canvas. Pill geometry. The system is the documentation.

Color

Surface & Ink

primary

#000000

surface-dark

#171717

canvas

#ffffff

surface-soft

#fafafa

hairline

#e5e5e5

Text

charcoal

#525252

body

#737373

mute

#a3a3a3

Traffic lights (terminal only)

terminal-red

#ff5f56

terminal-yellow

#ffbd2e

terminal-green

#27c93f

Typography

Token	Sample	Spec
display-xl	Get up and running.	Nunito · 36 / 500 · 0
display-lg	Pricing	Nunito · 30 / 500 · 0
heading-lg	Automate your work.	Nunito · 24 / 600 · 0
body-md	Run any model, locally or in the cloud.	Inter · 16 / 400 · 1.5
body-sm	Access larger models on data-center-grade hardware.	Inter · 14 / 400 · 1.43
code-md	curl -fsSL https://ollama.com/install.sh \| sh	ui-monospace · 16 / 400
button-md	Download	Inter · 14 / 500 · 1

Radius

none

0px

sm · code chip

6px

8px

lg · cards

12px

full · pills

9999px

Components

Buttons

Download Sign in

Install snippet