Repository navigation
Releases
Releases · qiaozy-16/dsh-vhcu
v1.0.1 - Platform Abstraction & Standalone Packaging
DSH-VHCU v1.0.1 Patch Release
This update is a milestone for the underlying architecture. We have completed the platform abstraction layer refactoring and standalone packaging loop, making it a true sandboxed CUA plugin.
🚀 Core Features
- Zero-Dependency: Standalone packaging via PyInstaller. No Python environment configuration required. Unzip and run.
- High-Priority Interrupt (HTTP /interrupt): Introduces an out-of-band interrupt signal. When "Soft Pause" (user mouse takeover) or "Corner Emergency Stop" occurs, it interrupts the underlying Python macro execution or visual inference in milliseconds.
- Absolute Path Sandboxing: The Node layer injects absolute paths for
models/macros/logsvia environment variables. Data is never polluted, and paths never break. - DriverFactory: Input, capture, and audio interfaces are fully decoupled, paving the way for future macOS adaptation.
- Dependency Slimming: Removed bloated third-party libraries like
torchandpandasto keep the host lightweight.
📦 Quick Start (Windows)
- Download Full Package: Go to Assets below and download
dsh-vhcu.zip(includes source code and pre-compiled local engine). - Extract: Extract it to a permanent directory (e.g.,
C:\dsh-vhcu). - Register Plugin: Open a terminal (PowerShell) in that directory and run the installation script once:
node scripts/install.mjs - Restart DSH: Completely restart DeepSeek Harness. You can now see and enable VHCU in the plugin list.
Note: The unpacked size is about 238MB as it contains the complete vision and audio engine. It runs purely locally with absolute safety.
vhcu v1.0.0Models
VHCU v1.0.0 Offline Models
This is the local speech recognition model (faster-whisper-base) required by the VHCU Hear module.
How to use:
- Download the
models.zipfile below. - Extract
models.zipinto the root directory of the VHCU plugin. (Ensure the extracted path isdsh-vhcu/models/faster-whisper-base). - Restart DSH, open the VHCU sidebar panel, and click "Enable Hear".
Note: This model is sourced from
SYSTRAN/faster-whisperand is licensed under the MIT Open Source License.