AI-OS (Second-Brain)

myos/product-vision.md

MyOS — the product

Reframed 2026-07-01. MyOS is a product, not just Arijit's personal setup. It is a multi-tenant platform for running a portfolio of projects: tracking them, executing on them, and showing their status. Arijit is dogfooding it as tenant 0. This doc is the product home; it will grow as the idea firms up. Things marked FORMING are still being thought through, do not treat as locked.

What MyOS is (one line)

A multi-tenant operating system for running your projects: it tracks every project, executes work on them (via an AI engine), and shows their live status.

The core framing (DECIDED 2026-07-01)

  • MyOS = the product. A multi-tenant platform, intended to be sellable to other customers.
  • ArijitOS = tenant 0. Arijit's own instance, the first customer, where we prove it by using it. His projects, his self-model (Operating-Principles, About-Arijit), his tracker (My-Projects) all live as tenant-0 data.
  • Other customers = other tenants. They plug their own projects/tools in and out. So every MyOS component must be tenant-scoped (nothing hardcoded to Arijit).
  • Hermes = the engine. The brain, executioner, and muscle of MyOS. Built so it can be productised. (Hermes already powers PRISM/Cassandra; MyOS reuses it as its execution core.)

Components (product features — specs in My-OS-Specs)

  • Project tracker — durable cross-thread, cross-tool record of every project (what / value prop / customer / stage / health / next + dated progress log). The tracker skill is run as a Claude Cowork routine (scheduled) so it stays current on its own.
  • Dashboard — visual/graphical project status. FORMING: frontend is undecided (see open decisions).
  • Persist integration — saving state also updates the tracker and shows the status board.
  • Cross-project + cross-tool awareness — knows where every project lives and which tool it runs in (Claude vs Codex) so parallel work never collides.
  • Housekeeping standards — repo structure (frontend/backend split), CLAUDE.md hygiene, code-health audits.

Open decisions (FORMING — do not lock)

  • Frontend: Claude Cowork vs custom build. Can Cowork show live dashboards and be the MyOS face, or does Arijit build a custom frontend (leaning custom, since MyOS runs on Hermes)? PARKED, Arijit will return to it. (Do not assert Cowork's dashboard capability without checking.)
  • Product name. "MyOS" is the working product name. Arijit mentioned "Hum" (unclear if a codename or a voice artifact) — confirm.
  • Multi-tenant architecture. Not yet designed. Near-term: build for tenant 0 (ArijitOS), but keep the seams tenant-ready (no Arijit-hardcoding), so multi-tenancy is a later refactor, not a rewrite.

Partner note (recommendation, not a yes)

Build for ONE tenant first (ArijitOS) and make it genuinely useful before investing in multi-tenant plumbing. Keep the data tenant-scoped from day one so productising later is a refactor, not a rebuild. Prove value on yourself; sell once it's real.

Pointers

  • Tenant-0 self-model + projects: Operating-Principles, About-Arijit, My-Projects.
  • Component specs: My-OS-Specs.
  • Engine: Hermes (see PRISM/Chowmes-PRISM for the current deployment pattern).
  • Hermes knowledge base (deep architecture, from source): HERMES-ARCHITECTURE-DEEP (1,132 lines, file:line cited — how the engine actually works + AIOS-fit + the hermes-0day sandbox constraint). Supporting: hermes/hermes-source-map.md, hermes/hermes-docs-overview.md, hermes/INBOUND-peer-session-teardown-2026-07-04.md. Canonical location: Projects/MyOS/hermes/.