Skip to main content
OpenWhispr’s AI processing feature lets you give instructions to an AI agent during dictation to automatically format, enhance, or transform your text.

​
How It Works

1

Name Your Agent

During setup, you choose a name for your AI assistant (e.g., “Assistant”, “Jarvis”, “Alex”)
2

Address Your Agent

Start your dictation with “Hey [AgentName],” followed by your instruction
3

AI Processes

The AI understands your request and formats/transforms your text accordingly
4

Agent Name Removed

The final output has the agent name reference removed automatically

​
Naming Your Agent

​
During Onboarding

When you first set up OpenWhispr, step 6 of 8 asks you to name your agent:
  • Choose a name that’s easy to say and memorable
  • Common choices: “Assistant”, “Jarvis”, “Claude”, “GPT”, “Alex”, “Sam”
  • The name helps OpenWhispr distinguish between regular dictation and AI commands

​
Changing Your Agent Name

You can change your agent’s name anytime:
  1. Open Settings → Intelligence
  2. Find Agent Name field
  3. Enter your new name
  4. Click Save
Your agent name is automatically added to your custom dictionary to ensure it’s always recognized correctly during transcription.

​
Example Commands

Format as Email

Input: “Hey Assistant, format this as a professional email. Thanks for the meeting today, let’s schedule a follow-up next week.”Output: Professional email format with greeting, body, and signature

Create Bullet List

Input: “Hey Claude, convert this to bullet points. We discussed the new feature, timeline concerns, and budget approval.”Output: Formatted bullet list with proper structure

Make Professional

Input: “Hey Jarvis, make this more formal. The thing is broken and we need to fix it ASAP.”Output: “The system is experiencing issues and requires immediate attention.”

Summarize

Input: “Hey GPT, summarize this meeting. [Long rambling notes about project status, team updates, action items]”Output: Concise summary with key points and action items

​
AI Provider Options

OpenWhispr supports multiple AI providers with different models optimized for various use cases.

​
Cloud Providers

Models:
  • GPT-5.2 — Latest flagship reasoning model
  • GPT-5 Mini — Fast and cost-efficient
  • GPT-5 Nano — Ultra-fast, low latency
  • GPT-4.1 — Strong baseline with 1M context
  • GPT-4.1 Mini — Smaller GPT-4.1 variant
  • GPT-4.1 Nano — Lowest latency GPT-4.1
Best For: General-purpose text transformation, reasoning tasksSetup: Enter your OpenAI API key in Settings → Intelligence → AI Provider

​
Local Models

Run AI models entirely on your device for complete privacy:
Models:
  • Qwen3 8B (5.0GB) — Latest with thinking mode support ⭐ Recommended
  • Qwen3 8B Q5 (5.9GB) — Higher quality
  • Qwen3 4B (2.5GB) — Compact with reasoning
  • Qwen3 1.7B (1.8GB) — Small but capable
  • Qwen3 0.6B (0.6GB) — Tiny for edge devices
  • Qwen3 32B (19.8GB) — Most powerful
  • Qwen2.5 0.5B-7B — Various sizes for different needs
Best For: Reasoning tasks, thinking mode, balanced performance

​
Setting Up Local Models

1

Select Local Provider

Go to Settings → Intelligence → AI Provider and select Local
2

Choose Model Family

Select Qwen, Llama, Mistral, OpenAI OSS, or Gemma
3

Select Model Size

Choose a model based on your hardware:
  • 8GB RAM or less: Use models under 2GB (Llama 3.2 3B, Qwen3 1.7B, Gemma 3 1B)
  • 16GB RAM: Use models up to 5GB (Qwen3 8B, Mistral 7B)
  • 32GB+ RAM: Use larger models (Qwen3 32B, GPT-OSS 20B)
4

Download Model

Click Download Model and wait for completion (downloads from HuggingFace)
5

Start Using

Once downloaded, the model runs entirely on your device
Local models require significant RAM and processing power. Start with smaller models (under 2GB) to test performance before downloading larger ones.

​
Common Use Cases

Command: “Hey Assistant, write a professional email thanking the client for their business and proposing a follow-up meeting next week.”Output: Fully formatted email with subject line, greeting, body, call-to-action, and signature
Command: “Hey Claude, organize these meeting notes with attendees, discussion points, and action items. [Rambling notes]”Output: Structured notes with clear sections and formatting
Command: “Hey Jarvis, add professional comments to this code. function processData array map filter reduce return result”Output: Properly commented code snippet explaining each step
Command: “Hey GPT, make this friendlier. We regret to inform you that your request has been denied due to policy violations.”Output: “We appreciate you reaching out! Unfortunately, we’re unable to approve your request at this time due to our current policies. Let’s discuss alternative options that might work better.”
Command: “Hey Assistant, translate this to Spanish. The meeting is scheduled for next Tuesday at 3 PM.”Output: “La reunión está programada para el próximo martes a las 3 PM.”
Command: “Hey Claude, format as a numbered checklist. Review document, send to legal, get approval, publish.”Output:
1. [ ] Review document
2. [ ] Send to legal
3. [ ] Get approval
4. [ ] Publish

​
Regular Dictation vs AI Commands

OpenWhispr automatically detects the difference:
TypePatternProcessing
Regular DictationNo agent nameDirect transcription only
AI CommandStarts with “Hey [AgentName],“Transcription + AI processing
Examples:
  • ❌ Regular: “This is a test message” → Output: “This is a test message”
  • ✅ AI: “Hey Assistant, capitalize this. this is a test message” → Output: “This Is A Test Message”

​
Privacy Considerations

Cloud Models

Your dictation is sent to the AI provider’s API for processing. Check provider privacy policies.

Local Models

Your dictation never leaves your device. Complete privacy for sensitive content.
When using cloud providers, only the transcribed text is sent to the AI — your audio file is never uploaded to AI services.

​
Configuring AI Settings

​
Selecting Provider and Model

  1. Open Settings → Intelligence
  2. Choose AI Provider (OpenAI, Anthropic, Gemini, Groq, or Local)
  3. Enter API key (for cloud providers)
  4. Select AI Model from dropdown
  5. Click Save

​
Advanced: Prompt Studio

Customize the system prompt that instructs the AI how to process your commands:
  1. Go to Settings → Intelligence → Prompt Studio
  2. Edit the system prompt template
  3. Use {agentName} and {transcription} variables
  4. Test with sample input
  5. Save your custom prompt
Editing the system prompt is for advanced users. The default prompt is optimized for most use cases.

​
Troubleshooting

  1. Verify your agent name is correct in Settings
  2. Make sure you’re saying “Hey [AgentName],” at the start
  3. Check that your API key is valid (for cloud providers)
  4. For local models, ensure the model is fully downloaded
  1. Try a smaller model (Qwen3 1.7B, Llama 3.2 3B, Gemma 3 1B)
  2. Close other resource-intensive applications
  3. Ensure you have enough free RAM
  4. Consider upgrading to cloud processing for speed
  1. Be more specific with your command
  2. Try rephrasing your instruction
  3. Use examples: “Hey Assistant, format like this example: [example]”
  4. Check Prompt Studio for system prompt issues
  1. Verify the API key is correct
  2. Check that your API account has credits/is active
  3. Ensure the key has proper permissions
  4. Try regenerating the API key from provider dashboard

​
Tips for Best Results

Be Specific

Clear instructions get better results: “Make this formal” vs “Rewrite professionally with business tone”

Provide Context

Include relevant details: “Format as email to client about project delay” vs just “Format as email”

Use Examples

Show what you want: “List like this: 1. Item, 2. Item” for consistent formatting

Iterate

If output isn’t perfect, dictate a follow-up command: “Hey Assistant, make that more concise”

​
Next Steps

Custom Dictionary

Add your agent name and technical terms for better recognition

Notes System

Use AI actions to enhance your voice notes

Transcription Modes

Configure your transcription provider and models

Basic Dictation

Master the fundamentals of OpenWhispr dictation

Build docs developers (and LLMs) love

Get started for freeTalk to us