Estate setup · macrothought.org
A working agent estate installed in your environment, on your machines, under your own Anthropic account. Nothing hosted, nothing proxied, nothing of ours between you and your model. You get the method that runs this site's own estate.
What gets built
- Department memory for your actual projects: one domain per folder, one topic per note, a loaded index, confidence tags on every claim.
- The verification layer: a memory lint with a positive control (we plant a defect and your lint must catch it), plus an audit of the one check or gate you already rely on most.
- Routines: session-end capture and a weekly lint review that actually fire.
- Session hygiene: formatting and secret-scan hooks where your stack supports them.
- The public skill pack, installed and tuned to your vocabulary.
- A mini brain-map: a self-contained page that renders your vault as a radial map. No external requests, works offline, yours to keep.
Out of scope: rebuilding our showcase apps, dashboards, or hosting anything for you. You get the working method installed, not our products rebuilt.
How it runs
Fully async, over email. Purchase triggers an intake questionnaire (fifteen minutes). Then a tailored runbook drives the build: each phase ends with a piece of evidence you send back. Email rounds are answered within one business day; a typical setup completes in five to ten days. There are no calls to schedule.
What "done" means
Delivery closes when four checks pass on your machine, run by you, with evidence:
| check | pass | fail |
|---|---|---|
| Recall | five questions about your projects answered with citations to real notes | an answer with no note behind it |
| Lint control | a planted defect fires the lint, removing it quiets the lint | lint stays green with the plant in place |
| Routine | observed firing on its schedule | only works when invoked by hand |
| Map | your vault renders in your browser | blank page or missing departments |
Every check has a defined failure. A setup that cannot fail its acceptance is not accepted, it is assumed. This is the same discipline the verification department teaches.
Terms, posted
- Full refund on request any time before work starts (before we send your tailored runbook).
- 50% refund after the runbook and department design are delivered.
- No refund after the four acceptance checks pass on your machine.
- 30 days of async email support after acceptance, for anything the setup broke or missed.
- One person delivers this. Five slots exist at this price so quality holds.
What it requires from you
- Your own machines and your own Anthropic account (any paid plan).
- Claude Code installed and working.
- Fifteen minutes for the intake, and email responsiveness during the build.
Questions before buying: open an issue on the repo and ask in public. If the answer would change your decision, it belongs where the next person can read it.