Use Claude Code on your mineral

Where you talk to your assistant. The app is the window onto your mineral; Claude Code is the room you work in.

You will be able to connect Claude Code to your mineral and work there, with your whole brain in context.

Where you talk to your assistant. The app is the window onto your mineral; Claude Code is the room you work in. The app shows you what it knows and lets you switch things on. Claude Code is where you do the interview, ask it things, and have it draft. Same assistant, same brain, the whole of it in context.

That is Claude Code, connected to your mineral over SSH. Same assistant, same brain, much more room. SSH is not a side door here: your keys are the only identity your mineral has, so this route is the product's own front door with the panelling off.

Sign in once in the app's Terminal, then open your mineral in Claude Code: the connection name from Help, Add SSH connection, pick the folder called state, say hello.

When this is worth it#

Every day. The one thing you do in the app's Terminal instead is the first sign-in to Claude, because that signs in the mineral itself and Claude Code cannot do that for it. Beyond the everyday, this is also the place for:

  • building or editing skills rather than running them
  • restructuring your brain, where you want to see and move files
  • doing something long and iterative that a chat window makes tedious

Getting connected#

The connect guide lives in the app under Help. It is there rather than here because it hands you the exact host for your mineral, and a page that told you the general shape would just be a worse version of the thing that knows your address.

What the guide gives you is an SSH host entry. Once that is in place, opening Claude Code against your mineral is the same as opening it against any machine you have access to.

The short shape, so you know what you are in for: in Claude Code, open the Environment dropdown, choose "+ Add SSH connection", name it anything you like, and put your connection name in the SSH Host field exactly as the app shows it. Port and identity file stay empty; your key is already on this computer and the app finds it from the connection name. If this computer was set up through the app in the first place, the connection is usually already sitting in that dropdown, folder included, so check there before adding anything.

When Claude Code asks which folder to open, pick the folder named after your mineral (a mineral that has not named its folder yet calls it "state"). The first connection takes a minute while the app sets itself up on your mineral; after that it is quick. And the first thing to say is "introduce yourself": your assistant already knows who you are and takes it from there.

One distinction that cost people hours: connecting Claude Code signs in your laptop and reaches into your mineral over SSH. It does not sign the mineral itself in, it does not switch on scheduled jobs, and nothing it does makes the Overview's sign-in step go green. That step is the Terminal tab, once: open it and run claude. The create wizard's finish checklist offers exactly that under Connect Claude, so a mineral set up recently has usually done it on day one. Then live here.

Connections you make in here stay chats-only#

A service you connect from inside a Claude Code session works when you chat and is invisible to your scheduled jobs, which load their connections from a different place. This is not hidden from you: the Connections page lists such a service honestly, marked "in your chats only, not in scheduled jobs", and gives it a one-click "Use in jobs too" button that moves it where jobs can load it. The sign-in survives the move.

What is different on a mineral#

Two things surprise people.

Your brain is just files. Markdown, in a git repository, with wikilinks between pages. There is no database and no API. You can grep it, edit it in place, move pages around, and your assistant will read whatever it finds next time. The brain graph in the app draws your actual structure rather than a structure it expected, so you can see the effect of a reorganisation immediately.

Skills are files too. A skill is a markdown file with frontmatter describing when it applies. Writing a new one is writing a file. Editing one is editing a file. Nothing needs to be registered, compiled or deployed: it is present, so it works.

That is the whole reason this route exists. Once you can see that the product is files, the ceiling on what you can make it do stops being our imagination.

Your mineral describes itself#

At the top of the brain sits a page the mineral writes about itself: what it is, who holds it, which machines can sign in, what is connected, which skills are installed and where each came from, and its last few custody events. It is rewritten on every status update, so it can never be older than the last run, and your assistant reads it before answering questions about what this mineral is or can do. You can read it too. It is just a file, like everything else here.

Skills have an address as well: your own live under the brain's .claude/skills/<name>/, one folder per skill, which is what write your own skill walks you through. The mineral's software version is on the self-description page too, and in the app under Help, in the Mineral software fold.

The terminal in the app#

There is also a Terminal tab in the app: the short way in when you are already there, for a quick question or running a skill by name, and the place the first Claude sign-in happens.

Use it for the sign-in and for one-off asks. Use Claude Code for the interview and for everything after.

Three cautions#

You have real access. From a session on your mineral you can delete your own brain. Your mineral has a nightly backup and a repository in your own account (back up your mineral), and neither is a reason to be casual the day you decide to reorganise everything.

The machinery is not yours to edit. The engine, the scheduler and the app are maintained by us and replaced when the mineral updates, so a change you make there is a change that disappears on the next restart. Everything under your own brain and your own skills survives updates, by design: that boundary is what who edits what describes.

Your edits are pushed on the mineral's schedule, not yours. The brain goes to your own repository nightly, at 03:50 mineral time, so a long evening of terminal edits sits only on the mineral until then. Before a big reorganisation, ask your assistant to push the brain first; then the state you are about to change is already somewhere safe. Back up and restore is the full picture.