KonfAI Apps
A stable KonfAI workflow, packaged with its weights and configuration as an app. The same app runs from a local folder, a Hugging Face repository or an HTTP server; from its own one-line CLI, from the generic konfai-apps runtime, from 3D Slicer, or as a tool of the MCP server. Consuming a published workflow does not require authoring YAML; the complete configuration stays available when you need to inspect, evaluate, fine-tune or automate it.
pip install impact_synth_konfai
impact-synth-konfai synthesize MR -i input_mr.nii.gz -o output/
# the same packaged workflow through the generic app runtime
konfai-apps infer VBoussot/ImpactSynth:MR -i input_mr.nii.gz -o output/The published bundles
Each figure is the medium case (249 × 246 × 246 voxels) of the bundle's own small, medium and large table, measured on an RTX PRO 5000 with benchmarks/perf/bench_apps.py on 2026-09-09; the ratios against the original tools span the three cases. They are evidence of executable scale, not a cross-task leaderboard; the protocol and the harness are under benchmarks/.
| App | What it does | Medium case |
|---|---|---|
| TotalSegmentator-KonfAI | whole-body CT or MRI segmentation with TotalSegmentator models; CT: 117 labels, 5 models; MRI: 50 labels, 2 models | CT total: 17.7 s, 5.2 GB RAM, 15.4 GB VRAM; 1.8 to 3.9 times faster and 2.7 to 4.8 times less host RAM than the original |
| MRSegmentator-KonfAI | whole-body MRI segmentation with MRSegmentator models; 40 labels, 5-fold ensemble | 21 s, 5.3 GB RAM, 16.3 GB VRAM; 1.1 to 1.7 times faster and 1.4 to 5.4 times less host RAM than the original |
| ImpactSynth | synthetic CT generation with IMPACT-Synth models; three MR and CBCT to sCT variants, 2.5D UNet++, 5 models each | MR: 24.6 s, 2.7 GB RAM, 12.8 GB VRAM |
| ImpactSeg | multimodal anatomical segmentation with IMPACT-Seg models; one model segments 11 structures from CT, MRI or CBCT | 3.3 s, 1.7 GB RAM, 3.9 GB VRAM |
| IMPACT-Reg | multimodal medical image registration with IMPACT-Reg presets; 13 presets across elastix with IMPACT, ConvexAdam and FireANTs | FireANTs_SyN: 108 s, 6.3 GB RAM, 16.0 GB VRAM |
The runtime: konfai-apps
konfai-apps is the standalone package for packaged workflows: local inference and evaluation, reusable app packaging, remote execution through a FastAPI server, integration in external tools such as 3D Slicer. It ships three interfaces: the konfai-apps command for local or remote execution, konfai-apps-server for hosting apps behind an HTTP API, and the Python API under konfai_apps. It depends on konfai and is versioned and tested as a separate runtime surface.
python -m pip install konfai-apps # run apps, locally or against a server
python -m pip install "konfai-apps[server]" # host apps behind an HTTP APIIn 3D Slicer: SlicerKonfAI runs segmentation, synthesis, evaluation and uncertainty apps; SlicerImpactReg runs the IMPACT-Reg presets and shows the registration results.
Guides: Ship an App, the CLI reference, the HTTP server API. Source: konfai-apps/ and apps/.