Skip to main content
SlasshyWispr supports flexible model configuration for speech-to-text (STT) and AI inference. You can run entirely online, entirely offline, or in hybrid mode with independent runtime selection for each component.

​
Runtime Mode Selection

​
sttRuntimeMode
RuntimeMode
default:"online"
Speech-to-text runtime modeChoose between:
  • online - Use cloud-based STT service via API
  • local - Use local STT models (Parakeet)
​
aiRuntimeMode
RuntimeMode
default:"online"
AI inference runtime modeChoose between:
  • online - Use cloud-based AI service via API
  • local - Use local AI models via Ollama
​
runtimeMode
RuntimeMode
default:"online"
Legacy runtime mode (applies to both STT and AI)This is maintained for backward compatibility. Use sttRuntimeMode and aiRuntimeMode for independent control.

​
Online Model Settings

Configure cloud-based model providers for online mode.
​
apiBaseUrl
string
default:""
API base URL for online model providerExamples:
  • OpenAI: https://api.openai.com/v1
  • Custom provider endpoint
Leave empty to use default provider.
​
apiKey
string
default:""
required
API authentication key for online services
Keep your API key secure. Enable rememberApiKey to persist it across sessions.
​
rememberApiKey
boolean
default:"false"
Persist API key in settings storageWhen enabled, your API key is saved locally and restored on app restart.
​
sttModelName
string
default:""
Model name for online speech-to-textExamples:
  • whisper-1
  • Provider-specific model identifier
​
aiModelName
string
default:""
Model name for online AI inferenceExamples:
  • gpt-4
  • gpt-3.5-turbo
  • claude-3-opus-20240229

​
Local Model Settings

Configure offline models for local inference.

​
Local STT Models

​
localSttModel
string
default:""
Local speech-to-text model selectionAvailable models:
  • Whisper Turbo (openai/whisper-large-v3-turbo) - 1.6 GB - Fastest large model
  • Whisper Large v3 (openai/whisper-large-v3) - 1.1 GB - Most accurate
  • Whisper Medium (openai/whisper-medium) - 492 MB - Balanced
  • Whisper Small (openai/whisper-small) - 487 MB - Lightweight
  • SenseVoice (FunAudioLLM/SenseVoiceSmall) - 160 MB - Compact option
  • Moonshine Base (UsefulSensors/moonshine-base) - 58 MB - Ultra-lightweight
Models must be downloaded before use. SlasshyWispr will guide you through the download process in Settings > Models.

​
Local AI Models (Ollama)

​
localOllamaBaseUrl
string
default:"http://127.0.0.1:11434"
Ollama service base URLDefault points to local Ollama instance. Change if running Ollama on a different host or port.
​
localOllamaModel
string
default:""
Ollama model name for local AI inferenceExamples:
  • llama2
  • mistral
  • codellama
  • phi3
Ollama must be installed and running separately. Pull models using ollama pull <model-name>.

​
Setup Guide

  1. Open Settings > Models
  2. Set STT Runtime to Online
  3. Set AI Runtime to Online
  4. Enter your API Base URL
  5. Enter your API Key
  6. Enable “Remember API Key” (optional)
  7. Specify STT and AI model names
  8. Test with a quick dictation

​
Best Practices

  • Privacy-first: Use local mode to keep all data on-device
  • Speed-first: Use online mode for fastest inference
  • Balanced: Use Parakeet v3 locally with online AI for fast STT and powerful responses
  • Hardware advisor: SlasshyWispr analyzes your CPU/GPU and suggests optimal local models
  • Incognito mode: Enable in Advanced Settings to prevent history logging regardless of runtime mode

Build docs developers (and LLMs) love

Get started for freeTalk to us