Marc Tristan VictoriaandGitHub
f228731141
* feat: add Llamanager plugin * fix image paths in README * docs: fix validation issues * fix: handle missing trailing separator in modelfile directory * fix: escape user input in shell commands * fix: disable modelfile editor when path is unset or misconfigured * fix: handle invalid Modelfile directory * docs: update README * fix: reload launcher model from config * docs: update demo assets
94 lines
4.7 KiB
Markdown
94 lines
4.7 KiB
Markdown
# Llamanager
|
|
|
|
> **A Noctalia v5 plugin for managing local Ollama models, Modelfiles, and runtime state.**
|
|
|
|
Llamanager provides a graphical frontend for [Ollama](https://ollama.com/) inside Noctalia. It wraps the Ollama CLI and HTTP API into a single panel, and allows model management, runtime inspection, model downloads, Modelfile editing, and launcher integration without requiring direct terminal interaction.
|
|
|
|
## Demo
|
|
|
|
|
|
|
|
|
|
## Plugin
|
|
|
|
| Field | Value |
|
|
| --------------- | ---------------------------------------------------------------------------------------- |
|
|
| ID | `marccvictoria/llamanager` |
|
|
| Entries | bar widget: `widget`; service: `registry`; launcher_provider: `launcher`; panel: `panel` |
|
|
| Launcher Prefix | `/ll` |
|
|
|
|
## Usage
|
|
|
|
Before using Llamanager, complete the following setup steps:
|
|
|
|
1. Install Ollama and ensure it is available on your `PATH`. See [Requirements](#requirements).
|
|
2. Open the plugin settings and configure:
|
|
- the Modelfile directory
|
|
- the preferred AI model used by the launcher
|
|
3. Add the Llamanager widget to your Noctalia bar.
|
|
|
|
## Requirements
|
|
|
|
- **Noctalia v5**
|
|
- `ollama` available on your `PATH`: required for model management, downloads, launching models, and Modelfile operations.
|
|
|
|
Verify the installation:
|
|
|
|
```bash
|
|
ollama --version
|
|
```
|
|
|
|
## Settings
|
|
|
|
| Setting | Type | Description |
|
|
| ---------------- | -------- | ------------------------------------- |
|
|
| `modelfile_path` | `folder` | Directory containing user Modelfiles. |
|
|
| `launcher` | `string` | Model used by the launcher. |
|
|
|
|
## Implementation
|
|
|
|
### Panel
|
|
|
|
`panel.luau` renders whichever view is currently active. User interactions dispatch commands through `llamanager.nextCommand`; the service performs the requested operation, updates state, and the panel re-renders.
|
|
|
|
`registry.luau` is the headless service responsible for interacting with Ollama. Filesystem operation, CLI invocation, and HTTP request originates here. It discovers installed models through `ollama list`, queries runtime information from the Ollama HTTP API, manages Modelfiles on disk, downloads models through `/api/pull`, and executes `ollama create` and `ollama rm` when building or deleting models.
|
|
|
|
**Model List**. The list displays every model currently installed in Ollama. To launch a model, select it from the model selector and click the launch button. This executes `ollama run <model>` and opens an interactive session in a new terminal window.
|
|
|
|
**Model Downloader**. Downloads models directly through Ollama's HTTP API. Downloads execute using the streaming `/api/pull` endpoint and automatically refresh the model library after completion.
|
|
|
|
**Modelfile Editor**. Modelfiles are ordinary text files stored inside the configured Modelfile directory, with the suffix `.modelfile`.
|
|
|
|
- **Load into Ollama**. Builds the modelfile into an ollama model by executing ollama create.
|
|
- **Delete model/modelfile**. Deleting a modelfile also deletes the model in ollama using `ollama rm`. In case the modelfile has not been loaded yet to ollama, it will only delete the modelfile. When deleting a model, it is recommended to include their tag, e.g. `qwen3:latest`.
|
|
- **Create Modelfile**. Creating a Modelfile requires the Name field to be populated. This only creates the Modelfile on the directory and does not register it with Ollama. To make the model available in Ollama, use _Load into Ollama_ after creating or editing the Modelfile.
|
|
- **Edit Modelfile**. Editing a Modelfile also requires the Name field to be populated. Saving changes only updates the Modelfile on the directory; it does not update the existing Ollama model. After modifying the Modelfile, use _Load into Ollama_ to rebuild and apply the changes.
|
|
|
|
**Runtime Dashboard**
|
|
|
|
Queries `http://localhost:11434/api/ps` and displays:
|
|
|
|
- Running models
|
|
- Parameter size
|
|
- Quantization level
|
|
- Context length
|
|
- Expiration time
|
|
- Ollama version
|
|
- API connectivity
|
|
|
|
### Launcher
|
|
|
|
The launcher entry `llamanger.luau` is independent from the panel. Queries submitted through `/ll` execute the configured model directly with `ollama run`.
|
|
|
|
The launcher provides quick model execution through `/ll <question> //`. Append `//` to send, the model response can be seen through notification and can be copied in your clipboard by pressing Enter key.
|
|
|
|
## IPC
|
|
|
|
```
|
|
noctalia msg panel-toggle marccvictoria/llamanager:panel
|
|
```
|
|
|
|
## License
|
|
|
|
[MIT](LICENSE)
|