Default mode
Local-first
Private Windows Dictation
Maximus Prime Software
PrimeDictate
Speak naturally. Write anywhere.
A system-wide voice workspace that turns speech into polished text with verified local GPU acceleration and an optional, independently configured AI refinement stage.
Local engines
CPU · CUDA · Vulkan
Cloud STT
3 providers
Access
Global hotkey
Built for daily flow
Dictate quickly without losing context
Choose Toggle or Push-to-Talk, watch the unobtrusive floating control, and keep working while transcription finalizes safely in the background.
Live hotkey validation and state-aware tray controls
Adaptive voice detection for quiet and short speech
Timed TXT, SRT, VTT, and JSON media exports
Searchable local history and privacy-safe diagnostics
Two-stage architecture
Transcription and refinement stay independent
PrimeDictate does not lock speech recognition and text processing to one vendor. Choose the best engine for each stage—or keep the complete workflow on your device.
Capture
Start from any Windows application with a configurable toggle or hold-to-talk shortcut.
Transcribe
Use verified Vulkan, NVIDIA CUDA, local CPU, or an explicitly selected cloud STT model.
Refine & deliver
Apply local rules or a separate AI model, then safely return the text to the original window.
Flexible inference
Use the hardware and provider that fit the moment
Local engines prioritize privacy and predictable cost. Cloud engines remain available when portability or model capability matters more.
01
VerifiedAMD / Vulkan
Bundled whisper.cpp v1.9.2 runtime with SHA-256 integrity checks and runtime GPU detection.
02
Local GPUNVIDIA CUDA
faster-whisper acceleration through CUDA and cuDNN for supported NVIDIA hardware.
03
PrivateMulti-core CPU
Local int8 inference for Windows systems without a compatible discrete GPU.
04
OptionalCloud STT
Explicit Groq, OpenAI Transcribe, or Gemini Audio selection with no silent upload path.
Core Capabilities
A complete dictation workspace
01
Global shortcut
Configurable Push-to-Talk or Toggle dictation in any Windows application
02
Local acceleration
SHA-256-verified whisper.cpp Vulkan runtime for compatible AMD, Intel, and NVIDIA GPUs
03
Flexible engines
NVIDIA CUDA, local multi-core CPU, or optional Groq, OpenAI, and Gemini transcription
04
Independent transcription and post-transcription cleanup providers and models
05
Rule-based local cleanup plus optional Ollama, LM Studio, Gemini, Groq, OpenAI, and Grok processing
06
Safe active-window paste with optional clipboard restoration
07
Media workflow
Chunked, cancellable audio/video transcription with TXT, SRT, VTT, and JSON exports
08
Local-first defaults with explicit permission required for cloud fallback
09
API credentials protected by Windows Credential Manager
10
Searchable local transcript history that can be disabled
11
Desktop controls
Floating recording overlay, state-aware system tray menu, and startup integration
12
Quality and diagnostics
Adaptive voice detection, redacted rotating logs, and privacy-safe diagnostic bundles
Privacy boundary
Your microphone is local unless you decide otherwise
Local processing is the default. Cloud services are visible choices, credentials remain protected by Windows, and fallback uploads require explicit permission.
- ✓API keys are stored in Windows Credential Manager—not plaintext JSON.
- ✓Cloud fallback is disabled until the user explicitly enables it.
- ✓History can be disabled and remains under the current Windows profile.
- ✓Safe paste aborts when the original target window cannot be restored.
Product
Gallery
3 views · Swipe or use the controls
1 / 3
Technology and architecture
- Desktop interface
- Python 3.12 and PySide6
- AMD / cross-vendor GPU
- whisper.cpp v1.9.2 with GGML Vulkan
- NVIDIA GPU
- faster-whisper with CUDA / cuDNN
- Local CPU
- faster-whisper multi-core int8
- Cloud STT
- Groq, OpenAI Transcribe, and Google Gemini Audio
- Credential storage
- Windows Credential Manager
System requirements
- Windows 10 or Windows 11 (64-bit)
- Microphone for live dictation
- Current compatible driver when using CUDA or Vulkan acceleration
- Internet connection only for model downloads or optional cloud providers