Ollama
Ollama and contributors · Desktop chat, CLI and local API for running and managing supported AI models, with model-specific hardware requirements and optional cloud services.
About
Ollama provides a desktop chat application, command-line tools and a local API for running and managing supported AI models. Developers and people exploring local AI can use it to chat, prototype model-powered applications and connect compatible coding tools. This cask installs Ollama.app and its bundled ollama executable, rather than the separate binary-only distribution.
Features
- Chat with supported models through the desktop application or command line.
- Download, list, run and manage model variants from the Ollama library.
- Use supported vision, embedding, tool-calling or decision models according to each model's capabilities.
- Integrate applications through the local REST API and official Python/JavaScript libraries.
- Create model configurations with Modelfiles and import supported model formats.
- Connect supported external coding agents through documented launch integrations; those agents remain separate products.
Installation and first use
Install brew install --cask ollama-app, or download the official macOS package and place Ollama.app in Applications. Start the app, select an appropriate model and send a prompt. The cask links its bundled ollama command into Homebrew's bin directory. The standalone app may offer to create a CLI link in /usr/local/bin if it is missing. CLI users can run ollama to choose a model or integration, or ollama run <model> for a specific model. Initial model downloads require network access and additional disk space. Python and JavaScript libraries are optional integrations, not required to launch the packaged Mac app.
System requirements and hardware
The official Mac documentation and current cask require macOS Sonoma14 or newer. Apple M-series Macs support CPU and GPU execution; x86 Intel Macs use CPU execution according to the official Mac requirements. Version0.40.0 automatically uses MLX on Apple Silicon for model architectures supported by that runtime; this is not a claim that every model runs on MLX. Performance, memory and available acceleration depend on model size, context length and hardware. No single RAM minimum was confirmed for all models. Model files can occupy tens to hundreds of GB. The app archive is206,260,689 bytes, separate from downloaded models and installed storage.
Platforms, dependencies and conflicts
Ollama provides official macOS, Windows and Linux installation options and an official Docker image. This cask is the macOS desktop package; Linux or Windows installation and GPU requirements differ. Third-party mobile clients listed by the project are integrations, not an official mobile build of this app. The cask declares no extra formula dependency and explicitly conflicts with ollama-binary. A separate Homebrew formula, manually installed executable or another running server may also overlap with PATH or the local service port; inspect the selected installation rather than starting duplicate servers. The API ordinarily uses localhost:11434. A local integration needs a running server and a model compatible with the requested operation.
Models, privacy and cleanup
Local model execution and optional cloud execution are different workflows. A selected cloud model requires network access and its applicable account and service terms; do not assume every prompt remains on the device merely because the client is installed locally. Downloaded models have their own licenses and usage conditions, independent of Ollama's software license. Local model use can continue without downloading again once the required files are present, subject to the features being used. Optional Homebrew zap deletes ~/.ollama, which contains models and configuration, plus application support data, preferences and caches. Preserve required data first; removing models can require large downloads again.
Accounts, costs and license
Ollama's source software is MIT-licensed. Local execution does not require a paid Ollama subscription, but uses your own computing resources and separately licensed models. Optional cloud models and services require their relevant sign-in and may have plan or usage limits; consult current official cloud documentation for entitlement and pricing. Integrated agents and third-party providers retain their own fees and accounts. Installation is not a grant of unrestricted rights to all models in the library.
Official references
New What’s new in 0.40.1 From Homebrew commits
This release has no separate notes.