C V Tarun
Home

2026 · Product & systems

Memory OS

A desktop-first AI operating system for personal knowledge: capture, retrieval, and long-horizon context.

AI · Desktop · Memory

GitHub

Overview

Memory OS architectureCapture shellMemory layerindex · rank · citeLocal-first storeRetrieval / agents
Capture shell over a local-first memory layer used by retrieval and agents.

Problem

Personal knowledge is fragmented across notes, chats, and files. Agents need a durable memory layer that still feels like a product, not a database.

Architecture

  • Local-first store with sync boundaries for private data.
  • Background indexing pipeline for documents and conversation traces.
  • A thin shell UI focused on search, capture, and citation.

Tech choices

  • Local-first where possible

    Privacy and latency matter more than multiplayer collaboration for personal memory.

Challenges

  • Keeping retrieval trustworthy as the corpus grows.

Trade-offs

  • Product surface stays minimal so the memory layer can evolve.

Lessons

  • The OS metaphor only works if capture is frictionless.