Hyperfix

the computer for ADHD superhumans

The loop

This computer only does one thing, in three moves. Brainstorm — dump everything, unfiltered; nothing is judged here. Plan — cut one idea into steps, and give every step a check: the observable fact that proves it is done. A step without a check is a wish. Execute — exactly one step is THE ONE. The rest are deliberately out of focus.

Gates, and why a step needs one

A step is green when its checks pass, amber while it is the one you are on, and red when a check has failed. Nothing is "basically done". The point is not discipline for its own sake — it is that deciding whether something is finished is expensive, and a check decides it once, in advance, while you still have the context.

The brain — your key, your model

Press B anywhere, or the BRAIN chip in the composer. Paste an OpenAI, Anthropic or OpenRouter key, or point it at a Claude Code or Codex CLI you already pay for. The key is encrypted at rest on your own box and no request for it leaves this machine. There is no balance here and nothing is metered: your plan is flat, and the tokens are billed to you by whoever sold you the key.

Claude Code and Codex

If you already have a Claude Code or Codex subscription, select it in the brain instead of pasting a key. An ACP bridge lets this computer's agent drive an editor on your real machine, so the plan and the code stay in the same conversation.

The tool surface — your agents, these rules

The brain you bring does not live here, so the rails cannot live in a prompt. They live in tools. Turn the tool surface on in the brain panel (B, section 04) and point Claude Code, Codex or Zed at this computer: claude mcp add --transport http hyperfix …. Your agent then gets the same verbs the computer uses on you — it reads THE ONE before it answers, parks a new idea instead of chasing it, and cannot close a step whose gate is still red. It is refused, in writing, with the list of what is missing. The token is yours and rotates from the same panel; rotating cuts off every client at once.

Nags

When THE ONE stalls, this computer says so. There is exactly one outbound lane and it is a webhook you own — your own bot, your own channel, your own rules. Nothing here has an account on a messaging platform, which is why there is no list of them to connect.

Your plan

One plan, $10 a month, one computer, one address. There are no tiers, no credits, no add-ons and nothing to upgrade. Cancelling stops the container; your files are yours and exportable the whole time.

When something is broken

Start with the Terminal — this is a real Linux box and you have root on it. Service logs are under /dev/shm. If the shell itself is the thing that broke, this page is served by a different code path and will keep working.

What things are called

This computer names things after what they are for. The right-hand column is what the same surface is called elsewhere, in case you arrived knowing that.

Todaywhat is in front of you right now, and nothing elseHome
Filesthe workspace — a real directory on a real diskMy Files
Terminala root shell on your own machineTerminal
Voiceshow the agent speaks to youPersonas
Guardrailsstanding instructions it must not breakRules
Brainwhich model thinks, and on whose keyAI
Looktype, colour and motionUX
← back to the computer