ollama

The ollama resource installs Ollama, a runtime for running large language models locally. On macOS it is installed via Homebrew and started as a background service; on Linux the official install script is used, which registers a systemd service automatically.

Parameters:

• models: (array[string]) AI models to pull and keep installed. Model names match those listed in the Ollama library (e.g. "llama3.2", "mistral", "qwen2.5-coder:7b"). Codify adds models that are missing and removes models that are no longer listed.

Example usage:

Install Ollama with a single model

[
  {
    "type": "ollama",
    "models": ["llama3.2"]
  }
]

Install Ollama with multiple models

[
  {
    "type": "ollama",
    "models": ["llama3.2", "mistral", "qwen2.5-coder"]
  }
]

Install Ollama without pulling any models

[
  {
    "type": "ollama"
  }
]

Notes:

• On macOS, Homebrew must be installed before applying the ollama resource. The homebrew resource can install it.
• On Linux, the official install script (https://ollama.com/install.sh) requires curl and sudo privileges. The script creates an ollama system user and registers a systemd service.
• Models can be large (several gigabytes each). Make sure you have sufficient disk space before adding them to your configuration.
• To see available model names and tags, visit ollama.com/library or run ollama list after installation.
• Removing the ollama resource stops and uninstalls the Ollama binary but does not automatically remove downloaded model data (stored in ~/.ollama on macOS or /usr/share/ollama on Linux for system-level data).