🛠 Millfolio: Hybrid AI for Personal Data Protection
A developer has introduced millfolio — a project for working with personal data using a hybrid AI approach. Instead of sending files to a cloud model, the system sends a "program" (code) to the device, written by a frontier model (such as Claude) based on an anonymized data schema (metadata). The code itself is executed in a local sandbox on the device, where a local model indexes and processes the actual values. The project stack is built on the Mojo language.
🌍 It demonstrates an effective local-first AI architecture, where privacy is protected by separating planning (in the cloud based on the schema) from execution (locally on the data). The use of Mojo for high-performance inference on consumer hardware shows the potential of this new language for AI infrastructure.
👤 This is a new way to use powerful models like Claude to analyze personal notes or financial data without uploading the actual documents to the internet. Privacy is increased because only column names and data types are sent to the cloud, not the values themselves.
Source 1: https://millfolio.app/blog/send-the-program-to-your-data/