2025–
NivaOS — a privacy-first personal AI platform
One assistant, many capabilities — and a routing policy that decides what may leave the machine.
Context
NivaOS is Farhaan's own R&D platform: what does a trustworthy AI assistant look like when privacy is an architecture decision, not a checkbox? It runs daily as his personal assistant platform.
Problem
Cloud AI is capable but indiscreet; local AI is private but limited. Real use needs both — without the user having to manage the boundary themselves, message by message.
Solution
A modular "AI operating system": an assistant core (FastAPI, MongoDB/Redis/Qdrant, Telegram interface) with persistent memory and scheduling; capability modules instead of a monolith; an operations cockpit; n8n workflow automation as the assistant's hands — and a three-tier privacy policy that routes every task to a local model (LM Studio) or the cloud based on the sensitivity of the data involved.
Outcome
A working system that combines agent memory, scheduling, automation and sensitivity-based model routing — running every day. The same patterns are exactly what business AI integrations need: this is the "thinks in AI systems, not just chatbots" proof.