This is a real screenshot of Macrothought running — every note, project and tool one person owns, mapped as a living constellation and governed by a pipeline that verifies work before it ships.
Nothing on this page is a mockup. Every image is captured from the running system; every number is read from it.
Switch the map to TREE and each department grows upward from a shared root — a game project, an AI stack, a teaching archive — each one a distinct organism you can watch change week to week.
TREE layout, live capture. Left to right: a game in development · the AI stack · emulation · tools · NAS · editor · personal — rooted in one markdown vault.
Most AI workflows end when the model stops talking. Macrothought's orchestration layer is a deterministic pipeline that turns intent into verified, gated releases — a completed agent ≠ a completed requirement ≠ a shippable release.
No model in the loop. The requirement ledger, verification harness and release gates are plain state and plain code. Completion is provable, not asserted.
Cheap and idempotent. Local models classify, compile context and propose simple fixes — propose-only, never silently deciding. Runs on the desk.
Where judgment matters. Frontier agents take design, hard repair and adversarial audit. Everything they claim is re-verified by the spine.
The same pattern — one readable vault, explicit confidence, verified releases — fits anywhere knowledge decays faster than people can maintain it.
The machine view: the entire working environment as a constellation — amber marks what changed in the last 48 hours.
The brain travels; the hands stay home. The whole system mirrors to a small always-on server and is reachable from a phone — maps, stats, history, consoles — over a private network.
Remote work is reviewed like a pull request. Changes made away from the desk aren't auto-committed: they queue with a summary and file list, and land only on a one-click local review.
⇄ remote tab, live capture — a pending remote changeset and its commit history.
Markdown first. Databases, graph stores and vector memories were tried — and parked. The institutional memory is text a human can read, diff and version.
Confidence is explicit. Every claim is tagged [verified] [inferred] or [uncertain] — dated, and re-checked weekly.
Gap-aware recall. Answers say what's stale and what's missing. A memory system that implies completeness is lying to you.
No queue without a drain. Every automation that accumulates work has a routine that consumes it. Machinery that only grows is machinery that rots.
A walkthrough is the live system on a real screen — the sky, the tree, the gates — not a deck.
REQUEST A WALKTHROUGH