Every operator has a pile of small tasks not worth a hire — and not worth ignoring.

I help run a slice of Fueled, fulcrumm, a couple of media brands, and a personal life — five email accounts, half a dozen calendars, a different voice for each brand, and a stack of weekly mechanics that don't justify a hire but eat real hours. Triaging mail, drafting in the right voice, pulling meeting notes into the right project, keeping track of what I decided last week. None of it is hard. All of it compounds.

An AI assistant that knows your context can finally close that gap.

Off-the-shelf assistants don't close the gap because they don't retain the context. Every task starts with the same setup tax: which account, which voice, which project, which tools, what I decided last week. The cost of that re-explanation is what kills the leverage — and what keeps the minutiae sitting on my desk instead of getting handled.

A personal operating system built on Claude Code and Obsidian. Context — ventures, voices, tools, accounts, what I decided last week — lives in a git-synced repo the assistant reads before doing anything. The vault is the long-term memory; Claude Code is the hands; rules in CLAUDE.md keep it honest. Accessible from anywhere via Claude Code's remote interface — the same operator in my pocket as on my desk.

Context as a repo

Voice guides, venture briefs, tool configs, and operating rules live in version control. The assistant reads before it acts.

Obsidian as long-term memory

Knowledge accrues in a linked Markdown vault — searchable by me, readable by the assistant, owned by neither.

Tool-aware by default

Gmail across five accounts, the Fueled ERP, Granola, Strava, Hevy, Reminders — each scoped to the project that needs it, with guardrails for the destructive ones.

Anywhere, via Claude Code remote

The same assistant on a laptop, a phone, or a borrowed machine. No local install, no lost context.

Git-synced and auditable

Every action ends in a commit. Nothing local, nothing lost, every change reviewable after the fact.

The minutiae get handled. I can ask for an inbox triage, a brand draft, or a staffing read without re-explaining myself, and the output lands in the right voice, in the right folder, committed and pushed. The system gets sharper every week instead of starting over — and it's with me whether I'm at a laptop or on a phone in a cab.

1. 01The unlock isn't a smarter model — it's giving the model a place to remember. Context in a repo beats context in a prompt, every time.
2. 02Guardrails are part of the product. 'Never send emails, always draft' and 'never edit outside these folders' are what make it safe to give an assistant real keys.
3. 03An audience of one is a real constraint. PhilOS is unapologetically not a product — which is exactly why it can be sharpened around the actual work.