Join Vultr Community
Points, levels & badges
Enter challenges & submit builds
Register to events & meetups
Get your work featured
Built a voice-controlled screenshot memory + AI assistant using Vultr Serverless Inference
Hey everyone! Wanted to share something I recently built using Vultr after seeing a itneresting blog
I take hundreds of screenshots (errors, meeting slides, dashboards) and could never find them later. I also hated switching to ChatGPT every time I wanted to draft a reply.
So I built a single desktop tool that solves both. It automatically watches for new screenshots, sends them to NVIDIA Nemotron Omni via Vultr Serverless Inference, generates a plain-English description, and stores it in Vultr Vector Store for semantic search. Hold a hotkey, speak your request, and it captures the active window, transcribes with Groq Whisper, searches your screenshot memory or drafts a context-aware reply, then pastes it directly into the active app without changing focus.
Stack: Vultr Serverless Inference • Vultr Vector Store • Groq Whisper
Happy to answer questions if anyone wants to build something similar!
Demo: https://youtu.be/BkjOi1JCEMo
GitHub: https://github.com/Cappybara12/wispr-better
KGSB
2 comments · incl. Vultr Team2 Comments
2
Commented: 3 months ago
Very good use of Vultr Serverless Inference showcased in the project.
2
0
Commented: 2 hours ago
Really clever idea, I lose screenshots all the time. How well do the Nemotron descriptions work on code and error screenshots compared to slides? And how long does it take from pressing the hotkey to the paste?
0