Local LLManager
Local LLManager is a macOS GUI for Ollama, written in Swift and SwiftUI. It provides a neat, user-friendly interface for interacting with the Ollama service — no Terminal commands necessary (outside of setup).
I mostly wrote this because I would start Ollama and forget about it in the CLI and thought it would be an engaging way to learn Swift.
Note
This is my first Swift project — it's mostly a demo at this point!
- Native MacOS app
- Support for light/dark mode
- Uses both CLI and API calls to interact with Ollama:
- CLI for:
- Starting/stopping the Ollama service
- Checking service status
- API for:
- Managing installed models
- Retrieving model status and data
- CLI for:
- Looks pretty (to me!)
- Model management (current)
- Release a build
- Implement better visual cues for loading
- Add mock data for Xcode preview views
- Model downloading support
- Specify model run duration and temperature (e.g., via keep_alive and options)
This project follows the MVVM architecture pattern to help me write the code in a cleaner way with a separation of responsibilities. Below is a breakdown of the core design decisions and how different pieces of functionality are implemented.
Service management is performed via CLI for simplicity and reliability.
I use pgrep to determine whether the Ollama service is currently running. This approach is fast and avoids the need for deeper system integration or dependencies.
Starting and stopping the service is also handled through CLI commands. The Ollama service cannot be started or stopped via. API.
All model-related functionality is implemented via the Ollama API. This includes:
- Listing installed models
- Listing running models
- Removing models
By isolating all API interaction in the ViewModel layer, the UI remains declarative and reactive.
The application refreshes the list of available models every 15 seconds using a scheduled task. This interval strikes a balance between staying up-to-date and minimizing unnecessary resource use. I opted for this polling-based approach rather than setting up notifications or event-driven updates due to the limited nature of Ollama’s current API.
All operations, both CLI and API, are executed asynchronously. This avoids UI stalls and keeps the interface responsive, particularly during potentially long-running tasks like querying models or starting the service.
- Ollama CLI installed via. Homebrew