Shadow Claude v0.2.0: GPU Acceleration & Model Choice
The biggest update since launch. Choose your Whisper model, get GPU-accelerated transcription, and never miss a question again.
What's New
- → NVIDIA GPU acceleration -10x faster transcription
- → Choose your model: Base, Small, or Medium Whisper
- → In-app model download with GPU detection and smart recommendations
- → Smarter question detection that works with any audio source
- → Background token refresh -no more mid-interview auth failures
- → "Get Answer" button for instant manual triggers
Why v0.2.0 Changes Everything
Shadow Claude v0.1.x shipped with a single Whisper model running on CPU. It worked, but it was a compromise -the Base model is fast but makes more transcription errors. The Medium model is far more accurate, but too slow for CPU.
v0.2.0 solves this by adding GPU acceleration. With an NVIDIA GPU, the Medium model runs in 1.5 seconds -faster than Base on CPU. You get the best accuracy without sacrificing speed. And if you don't have a GPU, Base still works great on CPU.
What Changed
GPU-Accelerated Transcription
Before
CPU only. Base model took ~2s per 30s of audio.
After
NVIDIA CUDA acceleration. Medium model runs in ~1.5s on GPU -faster than Base on CPU, with much better accuracy.
10x faster transcription with the best model. Questions detected sooner, answers arrive faster.
Choose Your Model
Before
One model bundled with the app. No choice.
After
Three models available: Base (fast, CPU-friendly), Small (balanced), Medium (best accuracy). Pick what fits your hardware.
Users with powerful GPUs get premium accuracy. Users on laptops still get a smooth experience with Base or Small.
In-App Model Downloads
Before
148MB model bundled in installer. One size fits all.
After
Lightweight installer (~500MB with CUDA). Models downloaded on first run with progress bar. GPU auto-detected, model recommended.
Smaller installer, faster setup, right model for your hardware.
Smarter Question Detection
Before
Relied on audio silence detection only. Missed questions when interviewer had no clear pause.
After
Transcript-settled detection: notices when speech content stops changing, triggers classification automatically.
Works reliably with system audio, continuous speech, and interviewers who don't pause clearly.
Token Auto-Refresh
Before
Access token expired after 1 hour. App stopped working mid-interview.
After
Background refresh every 45 minutes. Session stays alive for the entire interview.
No more mid-interview authentication failures.
Get Answer Button
Before
Had to wait for automatic detection or use keyboard shortcut (Ctrl+Shift+A).
After
One-click 'Get Answer' button right in the transcript panel. Send whatever's been said to Claude instantly.
Manual fallback when automatic detection doesn't trigger. No keyboard shortcut to remember.
Choose Your Model
On first launch, Shadow Claude detects your GPU and recommends the best model. You can change models anytime in Settings.
| Model | Size | CPU Speed | GPU Speed | Accuracy | Best For |
|---|---|---|---|---|---|
| Base | ~142 MB | ~2s | ~0.3s | Good | Older hardware, laptops without GPU |
| Small | ~466 MB | ~5-7s | ~0.6s | Better | Most users with a GPU |
| Medium | ~1.5 GB | ~15-20s | ~1.5s | Best | Users with 4GB+ GPU VRAM |
GPU speeds measured on NVIDIA RTX 3070 (8GB). All models support 12+ languages.
How to Get Started
Download
Get the latest installer from the download page. The app includes CUDA support -no separate GPU driver install needed.
Choose Your Model
On first launch, Shadow Claude detects your GPU and suggests the right model. Download it with one click -progress bar shows the status.
Start Listening
Log in, hit Start Listening, and you're ready. The app transcribes your interviewer's audio, detects questions, and sends them to Claude for answers.
Upgrade to v0.2.0
GPU acceleration, model choice, and smarter detection. Free plan still included -no credit card required.