Launched this week
Jompter
Dictation App Design For Agentic Developers
11 followers
Dictation App Design For Agentic Developers
11 followers
Jompter turns your voice, screenshots, and screen recordings into one clean, AI-ready Markdown prompt for Claude Code, Cursor, and ChatGPT. Local Whisper + open models run on your Mac — offline, private, no per-token billing.
As you know, I'm a developer, and I wanted an app that makes it easier to prompt AI coding agents like Claude Code, Cursor, Cortex, Kimi K3, and similar tools.
The biggest problem I found was that the built-in voice input wasn't accurate enough, and none of these tools had proper screenshot or visual context management.
So I built an app that lets you record a voice prompt while capturing your screen. You can drag and drop screenshots, annotate them, and explain exactly what you want. This makes prompting AI agents much faster and more accurate.
Another important feature is full local AI support. I don't want my voice recordings or sensitive project information being sent to external servers. Unlike many existing solutions, everything runs locally on your computer, giving you better privacy and lower latency.
I also added real-time streaming transcription, so you can see your prompt being transcribed live as you speak.
One of my favorite features is Explain Mode. You simply hover over any text generated by an AI agent and click Explain. The AI then breaks it down into simple, easy-to-understand language. This is incredibly useful because AI-generated plans can sometimes be difficult to understand, and Explain Mode makes them much clearer.
Since everything runs locally, the app is fast, private, and seamless to use.
This is honestly the best developer tool I've built so far, and I have many more features planned that will make AI-assisted development even better.
If this sounds interesting, let me know—I'd love to hear your thoughts.
Finally something that makes my messy voice notes actually useful with Claude Code. The offline Whisper piece is a nice touch, ran it on a long screen recording and it came out clean.
offline-first approach is genuinely cool, especially the local whisper part. one thing though, would love to see a way to batch process multiple recordings or screenshots at once into a single markdown prompt. right now it seems like you have to do them one by one which would get tedious pretty fast. basically a queue or folder drop feature would make this way more useful for longer workflows
One thing that would make this a daily driver for me: a quick way to grab just the last 20 seconds of audio and screenshots as a single snippet, without re-transcribing the whole session. Basically a "capture from here" hotkey that bundles everything since the last prompt into one Markdown block. Would speed up iterating on long debugging sessions a lot.
Local Whisper running offline on my Mac is a nice touch, especially since I got tired of my screen recordings eating up tokens in Cursor. Works exactly as advertised.
The whisper transcription running locally on my MacBook was surprisingly fast and the markdown output dropped straight into Cursor without any cleanup. Wish I'd had this a month ago when I was juggling three screen recordings for a debugging session.
Finally something that handles the messy voice-to-prompt workflow without sending audio off to some random server. Ran a quick screen recording through it on my Mac and the markdown came out clean enough to paste straight into Cursor.