FAQ

Common questions about installing, using, updating, and removing the SI.

Do I need to install Python first?

No. If you already have Python 3.10 or newer, the SI uses it; otherwise it brings its own managed Python. Either way, don't install it by hand.

Does it change how I use Claude Code?

No. Same commands, same workflow. The SI loads itself and works above Claude, replacing nothing about how you already work.

Will reinstalling touch my config or memory?

No. The installer is idempotent, your memory is create-only, and it merges only its own hooks and permission rules into settings.json.

Does it cost more or burn tokens?

The opposite is the goal. Routine work is routed to cheaper or local models, and the powerful model is saved for the hard parts.

Do I need a team or multiple machines?

No. The SI is great solo. Federation across machines is optional and off by default.

How do I remove it?

Run ~/.si/uninstall.sh (preview first with --dry-run; on Windows use ~/.si/uninstall.ps1). It offers to back up your memory first and strips only its own entries from settings.

Does it work on Windows?

Yes. In PowerShell run irm https://si.higgsdyson.com/install.ps1 | iex. Windows also needs Git Bash or WSL present, because the SI's hooks are bash scripts. The .\install.ps1 and install.cmd scripts from a clone are for contributors only. One difference: automatic restart-on-reboot isn't set up on Windows yet, so the background process self-starts each session instead.

Is my memory private?

Yes. Memory is stored locally under ~/.si/ and isn't sent anywhere by default. It's shared only if you enable federation, and then only what you accept.

One feature is worth knowing about explicitly, because it reads more than the SI's own notes. The SI can learn background context about you — your setup, your tools, the constraints you work under — by reading your own past conversations on this machine. It is off unless you turn it on, it never sends anything anywhere, and when enabled it refuses in code to store health, financial, relationship or other people's identity details. Turn it on by adding {"capture": {"mine_transcripts": true}} to ~/.si/cognition.json. Anything it learns this way is written to a manifest you can read, and any single item can be deleted.

How do I update?

You don't have to — the SI keeps itself current. In the background it downloads and verifies a new release, then at the start of a session (never mid-task) asks before applying it. Say yes and it updates itself; your memory and settings are left untouched.

If you'd rather start it yourself, say so in plain language — update the SI. That is the whole instruction; you are describing an outcome, not remembering a command. Roll that back and what changed in the last update? work the same way. The /si:si-upgrade slash command still exists for when you want to be unambiguous or are scripting.

Installs made from a source clone update the usual way with git pull && ./install.sh --update.

What if the install fails partway?

Re-run it — it's always safe and idempotent. A clean finish is exit code 0 with SI INSTALL: GREEN. The SI ONLINE greeting isn't from the installer; it appears the next time you start Claude Code, when the SI itself loads.

DEGRADED is not a failure. If the installer finishes with SI INSTALL: DEGRADED it did apply — everything is installed, and one optional thing reported a problem it could work around. The warning tells you what: a common one is cloud model credentials that have expired, which the SI simply routes around. Nothing needs re-running. Only FAILED means the install did not land.

What is a submind?

A named helper agent spun up for one focused job. It works off to the side, returns only its conclusions, and keeps your main session clean.

What does SI ONLINE mean?

It's the greeting that confirms memory has loaded and the SI is running at the start of a session. If you don't see it, restart Claude Code — the SI only loads on a fresh session.

Higgs Dyson LLC · SI documentation