Skip to main content

​
What is Hybrid Mode

Hybrid mode allows SlasshyWispr to intelligently route voice processing tasks between online (cloud-based) and local (on-device) models. This gives you the flexibility to balance between performance, privacy, and cost based on your needs. With hybrid mode, you can configure:
  • STT (Speech-to-Text) to use online or local models
  • AI (Assistant) to use online or local models
Each component can be set independently, allowing for combinations like:
  • Online STT + Local AI
  • Local STT + Online AI
  • All online or all local

​
How Hybrid Routing Works

SlasshyWispr provides three runtime modes for each component:
1

Online Mode

Routes requests to cloud-based API providers. Requires API credentials and internet connection. Typically offers the best quality and fastest processing for complex tasks.
2

Local Mode

Routes requests to models running on your device. For STT, this uses downloaded Parakeet or Whisper models. For AI, this uses locally running Ollama models. Works completely offline.
3

Hybrid Configuration

You set the runtime mode independently for STT and AI, creating a hybrid setup that matches your workflow.

​
Component-Level Configuration

The app tracks separate runtime modes:
  • sttRuntimeMode: Controls speech-to-text processing (online | local)
  • aiRuntimeMode: Controls AI assistant responses (online | local)
  • runtimeMode: Legacy setting for overall mode preference

​
When to Use Hybrid Mode

Use local STT + local AI when handling sensitive information. All processing stays on your device with no data sent to external servers.Example: Medical dictation, legal notes, financial planning
Use online STT + local AI for fast transcription with privacy-conscious responses.Example: General productivity work where transcription speed matters but you want to keep AI reasoning local
Use online STT + online AI when you need the highest quality results and have reliable internet.Example: Professional content creation, complex research queries
Use local STT + local AI when working without internet access or with unreliable connectivity.Example: Travel, remote locations, air-gapped environments

​
Configuration for Hybrid

1

Open Settings

Navigate to Settings > Models in SlasshyWispr.
2

Configure STT Runtime

Choose your STT runtime mode:
  • Select Online to use cloud-based speech recognition
  • Select Offline to use local Parakeet or Whisper models
If you choose Offline, you’ll need to download a local STT model from the available options:
  • Parakeet v3 (478 MB) - Recommended
  • Parakeet v2 (473 MB)
  • Whisper models (487 MB - 1.6 GB)
  • Moonshine Base (58.0 MB)
  • SenseVoice (160 MB)
3

Configure AI Runtime

Choose your AI runtime mode:
  • Select Online to use cloud-based language models
  • Select Offline to use local Ollama models
If you choose Offline, ensure Ollama is installed and pull the models you want to use.
4

Set Up Credentials (if using Online)

If either component uses online mode:
  • Enter your API Base URL
  • Add your API Key
  • Specify model names for STT and/or AI
5

Configure Local Models (if using Offline)

If using local STT:
  • Download your preferred model from the STT model list
  • Wait for model to load (first load may take time)
If using local AI:
  • Set Local Ollama Base URL (default: http://127.0.0.1:11434)
  • Select or pull an Ollama model

​
Best Practices

Start with Online Mode: If you’re new to SlasshyWispr, start with online mode for both components to get the best initial experience, then experiment with local models once you’re comfortable.
Hardware Matters: Local STT models perform best on systems with:
  • NVIDIA GPU with sufficient VRAM for GPU acceleration
  • At least 8GB RAM for CPU-based inference
  • Multi-core processors for faster processing
Check Settings > Offline STT for hardware-specific model recommendations.
Network Dependency: Online modes require active internet. If your connection is unstable, consider using local modes to avoid interruptions during dictation.

​
Optimization Tips

  1. Model Selection: Choose smaller local models (like Moonshine Base) for speed, larger models (like Whisper Large) for accuracy
  2. Warmup Models: Local STT models load faster after first use. Consider warming up your model before important sessions
  3. API Costs: Monitor your online API usage and switch to local models for routine tasks to reduce costs
  4. Latency Monitoring: Check Settings > Pipeline to view real-time STT, AI, and TTS latencies to tune your configuration

​
Example Scenarios

​
Scenario 1: Developer Workflow

Configuration: Local STT + Online AI Why: Fast local transcription for code comments and documentation, with powerful online AI for complex code generation and technical queries. Settings: 
sttRuntimeMode: "local"
localSttModel: "nvidia/parakeet-tdt-0.6b-v3"
aiRuntimeMode: "online"
aiModelName: "gpt-4" (or your preferred model)

​
Scenario 2: Privacy-Conscious Professional

Configuration: Local STT + Local AI Why: Complete offline operation, no data leaves your device. Perfect for confidential work. Settings: 
sttRuntimeMode: "local"
localSttModel: "nvidia/parakeet-tdt-0.6b-v3"
aiRuntimeMode: "local"
localOllamaModel: "llama3.2"
localOllamaBaseUrl: "http://127.0.0.1:11434"

​
Scenario 3: Content Creator

Configuration: Online STT + Online AI Why: Maximum quality for professional content creation with advanced language processing. Settings: 
sttRuntimeMode: "online"
sttModelName: "whisper-1"
aiRuntimeMode: "online"
aiModelName: "gpt-4"

​
Scenario 4: Mobile/Offline Worker

Configuration: Local STT + Local AI Why: Work anywhere without internet dependency. Settings: 
sttRuntimeMode: "local"
localSttModel: "UsefulSensors/moonshine-base" (smallest)
aiRuntimeMode: "local"
localOllamaModel: "llama3.2:1b" (efficient)
Test different configurations to find what works best for your hardware and workflow. You can switch between modes anytime without losing your settings.

Build docs developers (and LLMs) love

Get started for freeTalk to us