KonfAI Apps¶
KonfAI Apps are the deployment and reuse layer of the framework.
Where low-level KonfAI workflows are designed directly through YAML files and local Python modules, a KonfAI App bundles those assets into a reusable package that can run through:
konfai-appson the command linePython via
konfai_apps.KonfAIAppa remote FastAPI server via
konfai-apps-serverclients such as 3D Slicer integrations
What an app contains¶
A KonfAI App repository is recognized by the presence of an app.json file.
Typical contents are:
my_app/
├── app.json
├── Prediction.yml
├── Evaluation.yml
├── Uncertainty.yml
└── checkpoint.pt
Depending on the app, some files are optional:
Prediction.ymlis the core inference entrypointEvaluation.ymlis needed forevalUncertainty.ymlis needed foruncertaintyfine-tuning relies on training assets and checkpoint files that live next to the app
Local and remote usage¶
The same app command can run:
locally, through
KonfAIAppremotely, through
KonfAIAppClient, if--hostis provided
In remote mode, the client:
uploads files to the server
schedules the job
streams logs back to the client
returns a zipped result bundle
When to use apps¶
Use low-level YAML workflows when you are still designing or debugging a model.
Use KonfAI Apps when you want:
a stable inference interface
reusable packaging for a team
distribution through Hugging Face or a private repository
local and remote execution with the same user-facing command