Help
Using Thorim
A five-minute read that answers the questions every new customer asks. Skip ahead if a section looks familiar.
The 30-second mental model
Thorim is a one-time install that adds a kit of skills, agents, hooks, and slash commands into your AI coding tool. Once installed, the tool you already use (Claude Code, Cursor, Windsurf, Codex CLI, or Gemini CLI) gains a senior co-founder personality, project memory that persists between sessions, and 20 specialist sub-agents you can dispatch on demand.
You buy it once. You install it once per machine. It works on every project on that machine, forever. Updates are a single command. Uninstalling is a single command.
Where Thorim lives on your machine
Thorim doesn't create its own top-level folder. It installs into the existing config folder of whichever AI coding tool you use. For Claude Code that's ~/.claude/ on macOS and Linux, or C:\Users\You\.claude\ on Windows. For Cursor it's ~/.cursor/. Each AI tool has its own conventional config dir; Thorim plugs into whichever ones it finds.
Inside that folder, after install:
~/.claude/
├── .thorim-manifest.json # proof Thorim is installed
├── .thorim-version # current version (e.g. 0.1.0)
├── CLAUDE.md # the spine, auto-loaded every session
├── AGENTS.md # same content under the AGENTS.md name
├── agents/ # 20 specialist agents
├── commands/ # /hellothorim, /plan, /build, /doctor, ...
├── skills/ # 106 skills across 13 categories
├── hooks/ # 6 hook scripts (signal/voice/quality gates)
├── memory/ # memory template for new projects
└── settings.json # hook + permission wiringIf you ever wonder whether Thorim is installed, the answer is: does ~/.claude/.thorim-manifest.json exist? If yes, you're installed. If no, you're not.
One install per machine, not per cloud account
This trips up almost everyone the first time. Thorim is files on your machine. If you have a laptop, a VPS, and another VPS, that's three machines. Each needs its own install. There's no central account, no sync, no "I logged in on my laptop so my VPS also has it."
The phrase "installs globally" in the docs means "works across all projects on a single machine," not "works across all machines you own."
What "machine" means here: Every distinct OS instance counts. Your local laptop is one. Each VPS you SSH into is another. A Docker container is another. WSL is technically separate from native Windows. Install on each one where you want Thorim active.
Two-tier memory: global vs per-project
Thorim has two kinds of memory, and they live in different places.
| Tier | Where | What |
|---|---|---|
| Global | ~/.claude/ (or equivalent for your tool) |
The spine, agents, skills, commands, hooks. The same for every project. Installed once. |
| Per-project | ./thorim/ in each project folder |
Project-specific memory: what you're building, who it's for, recent decisions, blockers, open questions. |
The global tier is created by the installer. The per-project tier is created the first time you run /hellothorim inside a project. So you'll see a thorim/ folder appear at your project's root only after you've activated Thorim there.
What /hellothorim does
Open your AI coding tool, navigate to a project folder, and type /hellothorim. That's how every working session starts. The command does three different things depending on what it finds:
1. You've been here before
Thorim finds ./thorim/memory.md in this project and loads it. You get a short orientation: today's date, what you shipped last session, top one to three priorities, anything blocked. Two seconds, you're back in the headspace.
2. New project that already has substance
The project has files (a README, a package.json, git history, source folders) but no Thorim memory yet. Thorim scans the project, reads the recent files, and surfaces a draft summary: project name, stack, audience, what was being worked on. It writes the memory silently in the background, says "Thorim activated, I'm caught up," and surfaces the last thing you were working on with the previous AI before you activated Thorim.
3. Parent directory with multiple projects
You ran /hellothorim from your home folder or some other directory that holds several project subfolders. Thorim shows a numbered picker of the projects it found and asks which one you want to work on. Pick a number, Thorim treats that subfolder as the project and continues.
4. Genuinely empty directory
Nothing to scan, no project files anywhere. Thorim asks you in plain language what you're building, who it's for, where you're at. It uses your answer to set up the project memory.
Which AI tools work with Thorim
Thorim auto-detects and installs for any of these it finds on your machine:
- Claude Code (Anthropic). Primary integration. Detected at
~/.claude/. - Cursor. Detected at
~/.cursor/. - Windsurf (Codeium). Detected at
~/.codeium/windsurf/or~/.windsurf/. - Codex CLI (OpenAI). Detected at
~/.codex/. - Gemini CLI (Google). Detected at
~/.gemini/.
If you have more than one of these installed, the installer asks which ones you want Thorim set up for. You can pick all of them. Each tool gets its own Thorim install, independent of the others.
Updating Thorim
Just re-run the install command from your welcome email:
curl -fsSL https://install.thorim.io/<your-token> | bash(Or the PowerShell version on Windows.) The installer detects that Thorim is already there, silently refreshes every file to the latest version, and exits. No prompts, no questions, no opportunity to break anything. We call this self-healing: every install run leaves you with a working, current install regardless of the state you were in beforehand.
If you've added your own custom files to ~/.claude/ (your own commands, your own skills under different names), those stay put. Only Thorim's own files get refreshed.
One thing the installer can't do for you: refresh your already-open AI coding chat. When you install or update Thorim, the new commands only appear in chats opened after the install completes. Close any open Claude Code or Cursor chat and start a fresh one to pick up the new commands.
Uninstalling Thorim
The fastest way is the cleanup command:
npx thorim removeIt restores any backup files we made during install, removes everything Thorim added, and leaves your AI tool exactly the way we found it. Your project memory at ./thorim/ in each project folder is untouched.
If for some reason the command doesn't work, manual cleanup is also possible: delete the Thorim-owned files inside ~/.claude/ (everything listed in .thorim-manifest.json) and rename any .thorim-prev/ backup folder back to its original name to restore your pre-install state.
Common issues and fixes
/hellothorim doesn't autocomplete in my Claude Code chat
Your chat session was already open when you installed Thorim. Claude Code reads its command list once at session start. Close the chat and start a new one; the command will be available there.
Install failed on Windows with error "bash: not recognized"
You're in Command Prompt instead of PowerShell. The bash version of the install command won't work in cmd. Open PowerShell (Start menu, type "PowerShell"), and use the PowerShell-flavored command from your welcome email: iwr -useb <url>.ps1 | iex.
Install failed with "Node.js 20+ required"
Node 20 or newer is required. The installer can install it for you with one prompt. If you said no, install Node 20 from nodejs.org and re-run the install command.
I'm running as root on my VPS, is that a problem?
No, you'll see a yellow warning ("You're installing as root, that's fine on a single-user VPS") and the install proceeds. Many vibe coders run as root on their personal VPS. If your VPS is shared with other users, run as your normal user account instead.
The install said "Refreshed" but my new /hellothorim doesn't work
Almost certainly the chat-cache issue from above. Close every Claude Code or Cursor chat tab, then open a new one. /hellothorim should now be in the autocomplete list.
I ran the install in the wrong directory, did I break anything?
No. Thorim installs to your home directory's config folder (~/.claude/), not the directory you ran the command from. The working directory doesn't matter for install.
When something feels off, run /doctor
Two ways to get a clear status report:
- Inside any AI coder chat: type
/doctor - Outside, in a terminal:
curl -fsSL https://install.thorim.io/<your-token> | bash -s -- --doctor
You'll get a green-and-red report covering the environment (disk space, home directory, network, target writable) and the install itself (manifest, spine, all agents/commands/skills/hooks present and intact, all the way down to file shebangs being valid). If anything's red, the report tells you the suggested fix.
This is the diagnostic to paste into a support email if something's wrong. We can read it instantly without needing follow-up questions.
Still stuck? Talk to a human
If /doctor didn't sort it, or you're confused about something the docs don't cover, email thorimapp@gmail.com. You bought this product from one person, not a faceless company. We read every email.
Reply hours: Monday to Friday, 9am to 5pm. Outside those hours we'll get back to you the next working day.
What to include in a support email: the output of /doctor (or --doctor), your operating system, your AI coding tool, and what you were trying to do when it didn't work. That's usually enough to diagnose without back-and-forth.
One last thing
Thorim is a small product run by a small team. The docs you're reading right now were written by the same person who'll answer your email. If something's confusing on this page, that's our bug to fix, not your problem to work around. Send a note and we'll rewrite the section.