Skip to main content
Agent mode gives you a voice-controlled AI assistant. Name it, talk to it, and get streaming responses from your preferred AI provider.

​
Setup

  1. Go to Settings > Agent Mode and toggle it on
  2. Set a hotkey to open the agent overlay
  3. Choose your AI provider and model

​
Naming your agent

During onboarding (or in settings), give your agent a name — “Jarvis”, “Alex”, or anything you like. Then speak commands like:
  • “Hey Jarvis, make this more professional”
  • “Hey Jarvis, write a thank you email”
  • “Hey Jarvis, convert this to bullet points”
OpenWhispr detects when you’re addressing your agent versus dictating regular text, and removes the agent name from the final output.

​
Supported providers

ProviderModels
OpenAIGPT-5 and GPT-4.1 families
AnthropicClaude Fable, Opus, Sonnet, and Haiku families
GoogleGemini and Gemma models
GroqQwen, GPT-OSS, Llama, Compound, and Kimi models
OpenRouter300+ models from many providers, via a searchable, grouped picker
LocalQwen, LLaMA, Mistral, Gemma via llama.cpp
EnterpriseAWS Bedrock — see enterprise providers
Configure API keys in Settings under each provider section, or use OpenWhispr Cloud for instant access.

​
Agent tools

The agent can interact with your notes:
  • Search notes — finds relevant notes using semantic search
  • Create notes — saves information directly to your notes
  • Update notes — modifies existing notes based on your instructions

​
Chat overlay

The agent overlay is a resizable glassmorphism panel that appears over your screen. It supports:
  • Real-time streaming responses
  • Conversation history (saved and resumable)
  • Custom system prompts
  • Floating and sidebar modes in the note editor