Building a local-first AI system around real administrative work

LifeOS started with three ordinary questions: what changed, what needs attention and what should happen next?

Marc Thornton designed and built the system as a self-hosted personal intelligence platform rather than a generic chatbot. An Android companion, email, scanned documents and messages feed into structured records connected to the relevant people, tasks and cases.

The design deliberately keeps evidence close to the answer. Source material is retained so conclusions can be checked, and changes to important records require confirmation rather than allowing an AI system to modify them invisibly.

The current stack combines Python and FastAPI services, Docker, Qdrant vector search and local models running through Ollama.

The project is developed against real administrative cases rather than abstract demos, so day-to-day friction helps define priorities and acceptance criteria.

LifeOS remains a personal product-development and R&D programme, not a packaged client product. Its relevance to Digital SHIFT is the method behind it: start with a messy real workflow, decide what should be automated and what still needs human control, keep provenance intact, then build the smallest useful system around those constraints.