AGENT AS A SKILL
YunXiaoHe Agent-as-a-Skill
Add task planning and tool coordination to Codex, DSH or your own agent. Keep the model you already use.
Choose this productYH INTELLIGENCE / PRODUCTS
Products & Downloads
Install YunXiaoHe in your agent, or download a local app. Choose a product below to get started.
1. Choose a product
All three work with your own model. Choose a product to see its installation options.
SKILL + APP
YunXiaoHe - Data Governance Agent
Clean tables, run statistics and train RISA models on your computer. Save a model to use with new data.
Choose this productSKILL + APP
YunXiaoHe - Supply Chain Analysis Agent
Analyze inventory, materials and suppliers, compare scenarios and get purchasing suggestions with local Avalon tools.
Choose this productOther models and applications
Models and applications
YH / Model weights
Hugging FaceModel weights
RISA
Recaptured-image detection
RISA combines visual representations with edge and texture features to screen for images photographed from a display.
94.19%Group cross-validation accuracy
26downloads in the last 30 days
HF update · Details and contents
A research model for distinguishing direct photographs from images photographed again from a display. RISA reads frozen visual representations together with edge and texture features.
- Group cross-validation accuracy
- 94.19%
- Evaluation data
- DS-05 · 86 images / 12 content groups
Includes Safetensors weights, configuration, and evaluation notes. Inference takes vectors produced by the compatible feature pipeline.
YH / Windows application
Hugging FaceWindows application
PSI Compact
Local text-to-video generation
Use PSI staged execution to run LTX-Video 2B and generate video locally on a Windows PC.
6 GiBLowest verified VRAM
0downloads in the last 30 days
HF update · Details and contents
Run LTX-Video 2B on a Windows PC. PSI manages memory through staged execution, accepts a text prompt, and produces an MP4 locally. Prompts and videos stay on the device.
- Environment
- Windows 10/11 x64 · NVIDIA GPU
- Lowest verified VRAM
- 6 GiB
- Model and output
- BF16 · 704 × 448 / 5 seconds
Includes the installer, runtime components, and a SHA-256 checksum. Storage, driver requirements, and the LTXV license are documented in the release notes.
YH / Model weights
Hugging FaceModel weights
TinySeaState
LatentINT8 sea-state classification
A lightweight classifier for four sea states, with latent INT8 voting for edge deployment.
24,984Encoder parameters
0downloads in the last 30 days
HF update · Details and contents
A four-class sea-state classifier for ocean-surface images. A lightweight encoder produces spatial features, followed by a latent INT8 voting head designed for edge-device perception.
- Encoder parameters
- 24,984
- MU-SSiD official test accuracy
- 89.02% · 4,800 images
Includes Safetensors and PyTorch weights, preprocessing conventions, the class map, and evaluation results.
Files, licenses and evaluation methods are in the linked repositories. Earlier entries were checked on September 17, 2026; the Vision record was added on September 21, 2026. These are dated results, not new tests of the current download.
Some Hugging Face entries show downloads in the last 30 days, synchronized hourly. Counts do not represent unique users. See the definitions for models and datasets.