A Mac app that keeps memory across local Ollama models
This Mac app provides a graphical way to use Ollama without working through the terminal. It automatically finds installed models and can see model files on disk even when Ollama is not running. Its semi- carries some conversation context across model changes and is designed to repair itself if the stored memory becomes damaged.
A read-only adapter connects the app to Ollama without downloading, copying, or changing model files. Qwen3 4B is included so the app works immediately, but users can connect their own models. A accepts hands-free questions and speaks each sentence of the answer as it is produced.
The works only while the app window is open, and audio is not sent to an outside service. Sensitive information can be kept in an AES-256 encrypted vault.
Key points
- The app automatically detects installed Ollama models and lets users switch models during a conversation.
- Its read-only connection does not download, copy, or alter existing model files.
- Semi- carries conversation context between different models.
- Voice data stays local, and the activates only while the app window is open.
- An AES-256 encrypted vault protects sensitive information.