Hearsy LogoHearsy
LLM Engine
ummm, i think... we should... like, go now?
Analyzing Audio

From rough dictation to polished text

Optionally refine your transcriptions with AI. Remove filler words, fix grammar, summarize, or apply custom prompts — using Claude, OpenAI, or a fully local model.

Built-in enhancement templates

Choose a template and Hearsy automatically applies it to every transcription. See the transformation happen in real-time.

Remove filler words

Strips um, uh, like, you know, and other verbal fillers automatically.

Raw Input

So, um, I think that, uh, we should probably, like, schedule a meeting to, you know, discuss the project timeline.

Add punctuation

Inserts proper punctuation, capitalization, and fixes grammar on the fly.

Raw Input

the quarterly report shows strong growth in all regions especially asia pacific which exceeded targets by fifteen percent

Summarize

Condenses long dictations into clear, concise summaries.

Raw Input

We spent the entire morning discussing the various possibilities for the new product launch and after weighing all the pros and cons we finally decided on a digital-first approach with a phased rollout starting in Q2.

Fix grammar

Corrects grammatical errors, awkward phrasing, and sentence structure.

Raw Input

me and him was talking about how the project dont have enough resources and we needs to hire more peoples

Choose your AI engine

Hearsy supports three LLM providers for AI enhancement, each with different strengths. You bring your own API key for cloud providers, or use the bundled local model for fully private enhancement.

Switch between providers anytime in settings. Your API keys are stored securely in macOS Keychain — never transmitted to Hearsy's servers. The local Qwen model requires no API key and no internet connection.

All three providers work with every enhancement template and custom prompt. The difference is in quality, speed, and privacy tradeoffs.

Claude
Anthropic · Cloud API

Best writing quality and nuanced understanding

OpenAI
OpenAI · Cloud API

Fast, reliable processing with broad capabilities

Qwen 2.5 3B
Local via MLX · On-device

100% private — runs entirely on your Mac

How AI enhancement works

Enhancement is a single extra step in the dictation pipeline — completely optional.

1

Raw transcription

Your speech is transcribed locally by Parakeet or Whisper. This step always happens on-device.

2

LLM processing

The raw text is sent to your selected LLM with the active template or custom prompt. Local Qwen stays on-device; Claude/OpenAI use their APIs.

3

Enhanced text pasted

The refined text is placed on your clipboard and auto-pasted to the active app. The entire flow takes 1-3 seconds.

Custom Prompt

Rewrite this as a professional email. Keep the key points but make it concise, polite, and well-structured. Add a greeting and sign-off.

Input

hey so the deadline is moved to friday and we need the designs by wednesday can you make that work

Output

Hi team, The project deadline has been moved to Friday. Could you have the designs ready by Wednesday? Let me know if that timeline works. Best regards

Write your own prompts

Beyond the built-in templates, you can create custom prompts that transform your dictation in any way you need. Write a prompt once, and it's applied to every dictation automatically.

Use cases are limitless — rewrite as professional emails, extract action items from meeting notes, translate to another language, convert to bullet points, generate code comments, or match your company's writing style guide.

Custom prompts work with all three LLM providers. Save multiple prompts and switch between them based on what you're working on.

AI Enhancement FAQ

Smarter dictation, your way

Download Hearsy and let AI polish your words.

macOS 14+ · Apple Silicon · Claude, OpenAI, or local Qwen