Project write-up
Crypton Lite: a private AI assistant that runs on your laptop and listens to your voice
Crypton Lite is a local-first voice AI assistant that creates tasks from your documents with page citations and keeps your data on your laptop.
Quick facts
- Project
- Crypton Lite
- Built for
- CRYPTON: Speed of Thought (WeMakeDevs), 31 October 2026
- Idea
- A private assistant that runs locally and listens to your voice
- Local model
- A 4-bit 4B model that fits a 6 GB GPU
- Network used for
- Voice transcription (Wispr Flow) and live web research (Anakin.io)
What it does
You can say "Create preparation tasks based on the event agenda document" and a small local model reads your PDF, creates the tasks and answers with citations that point to the exact page. Say "Open calculator" and it opens it. Ask for "best laptop under 50k" and it researches the live web.
Other commands include adding or deleting tasks (deletes always ask for a Confirm click), saving a searchable note with "Remember: …", and opening an app, folder or website from a fixed safe list.
Why it exists
People capture tasks while their hands are busy, and cloud assistants send private documents away. Crypton Lite keeps the model, retrieval, tools and storage local, so your documents, tasks and database never leave the machine.
What stays local and what does not
The language model, document retrieval, tools and storage all run on the laptop. The network is used only for voice transcription and live web research.
Related work
Crypton Lite grew out of a larger idea I call Crypton, a personal AI operating system. It sits next to my other agentic and RAG projects on the portfolio.
Frequently asked questions
What is Crypton Lite?
A private AI assistant that runs on your own laptop and listens to your voice, built for CRYPTON: Speed of Thought.
Does Crypton Lite send my documents to the cloud?
No. Documents, tasks and the database stay on the machine. The network is used only for voice transcription and live web research.
What hardware does it need?
Its 4-bit 4B local model is designed to fit a 6 GB GPU.
Links
Work with me
I build agentic AI systems, RAG pipelines, automation and full-stack products. If you have something worth building, get in touch or email theeditornitish@gmail.com.