All projects

Personal AI · Memory · Agent orchestration

JARVIS · Personal intelligence

A self-hosted personal assistant connecting persistent memory, tool-using agents, voice, and home infrastructure into one system.

An intelligence that stays with you.

JARVIS is my personal intelligence system: a persistent assistant built around context, memory, and the ability to act. I built and deployed its first version on my home infrastructure, with a dedicated always-on host and a separate GPU machine for local inference and heavier processing.

The engineering spans a memory graph with semantic and full-text retrieval, a task planner, tool-using agents, and a traceable execution pipeline. Tasks have explicit states and retry budgets, so a failed action can be surfaced instead of disappearing into an endless loop.

I built voice components for speech recognition and synthesis, an Alexa skill backend, Telegram access, and a web control interface. The broader work also includes an iPhone-oriented PWA for chat and visibility into the server fleet.

The central design question is how personal context becomes useful action. Local models and stronger reasoning tools sit behind a common system, with deliberate boundaries around private information, tool access, and actions that need human judgment.