> ## Documentation Index > Fetch the complete documentation index at: https://docs.tryhoard.com/llms.txt > Use this file to discover all available pages before exploring further. # Talk to your store > Start asking your Hoard store questions in plain English from Claude, ChatGPT, or Codex. Begin with one safe question and work up to answers your spreadsheet can't give you. Once Hoard is connected to an AI assistant, you can ask your store questions the way you'd ask a sharp employee, "what sold best last week?", "which cards are way underpriced?", "what should I deal with today?", and get an answer pulled from your real, live Hoard data. The questions that used to mean exporting a CSV and building a pivot table are now one sentence. Some of them are genuinely faster to ask than to find by clicking around. This page walks you up to that in three gears, so by the end you trust it enough to lean on it. This is different from **Hoard Desktop**, the app that syncs your TCGplayer inventory in the background. The assistant talks to the data Hoard already has — it works from any browser, on any device, and your home computer doesn't need to be on. See the [FAQ](/ai-assistant-faq) if that distinction is fuzzy. ## What you'll need * A paid Hoard plan ([Gold, Platinum, or Diamond](https://www.tryhoard.com/pricing)) * An AI assistant account: Claude (claude.ai or desktop), ChatGPT (Plus, Pro, or Business), or the Codex CLI ## Connect one assistant Pick whichever you already use. Each guide is a few clicks and uses a secure sign-in handshake, so you're never pasting a password or an API key into the chat. If you can already hold a conversation with Claude or ChatGPT, you have everything you need — there's nothing new to learn on their side. claude.ai or the desktop app Plus, Pro, or Business The OpenAI Codex CLI The fastest path for Claude or ChatGPT: add a custom connector pointing at ```text theme={null} https://mcp.tryhoard.com/mcp ``` then sign in to Hoard in the window that pops up and click **Allow**. Full steps with screenshots are in the per-assistant guides above. ## Crawl: prove it can see your store Start with one question you already know the rough answer to: > "What's my total listed inventory value?" You have a number in your head for this. When the answer lands close to it, you've just confirmed the assistant is reading your real store, not guessing from thin air. Nothing changed, either — a question like this only looks, it can't touch a price. Two more in the same spirit if you want to be sure: * "How many cards do I have listed?" * "When did Hoard last sync?" That's the whole point of this first gear. You're not finding out whether the AI is smart. You're finding out whether it's actually looking at your store. ## Walk: get answers you'd normally export for Now ask the things you'd usually pull a CSV to find out: * "Break my inventory value down by game." * "Show me my 10 most valuable cards." * "What did I sell last month, after fees?" * "Who are my repeat customers?" Every one of these is a sentence here and a sort-and-filter chore in a spreadsheet. Keep checking the answers against what you already know. That's how the trust builds, and after a few you'll stop double-checking, which is the moment it starts saving you real time. The assistant answers from your **last-synced** data, not a live peek at TCGplayer. If the numbers feel stale, ask it to trigger a sync (or run one from your dashboard) and ask again. ## Run: ask things your spreadsheet can't This is where it pulls ahead of both the dashboard and a stack of exports. Ask one question that crosses several things at once: > "Which of my cards are more than 30% below market, have sold at least once in the last month, sorted by how much money I'm leaving on the table?" That's your prices, the live market, and your sales history, all in one sentence. In a spreadsheet that's an export, a couple of lookups, and a manual sort. Here it's a question, and you can keep shaping it without starting over: * "Now just the Magic ones." * "Ignore anything under \$2." * "Group it by set." That back-and-forth is the part the dashboard can't really do. Instead of pulling a fixed report and eyeballing it, you talk your way down to the exact cards that matter. And when you're ready to act on what you found, you get the one thing a spreadsheet will never give you: a safety net. Ask the assistant to actually change a price and it **previews first** — you see exactly what would happen, with a risk rating, before anything touches your live listings, and hard price floors sit underneath so a fat-fingered "set everything to a dollar" can't go through. Excel will happily let you wreck a column. This won't. Once it's part of your day, the standing question is just: > "What should I look at today?" ## Ground it in your own store first AI models know Magic well and Pokemon decently, but their general knowledge of newer games (Lorcana, One Piece, Riftbound, and so on) is thinner. The fix is the habit you've already built above: let it read your data before it reasons. Opening a chat with something like "I sell mostly Lorcana, here's a snapshot of my store" and letting it run a couple of reads first makes every answer after that sharper, because it's working from your real numbers instead of generic priors. More on this in [what your assistant can do](/agent-capabilities#setting-expectations-i-sell-lorcana-one-piece-riftbound). ## When you're ready to let it make changes Reading is the safe default, and the write side is **off until you turn it on**. Flip it on at [Settings → Assistants](https://www.tryhoard.com/settings/connected-apps), then walk through your first real change with the preview step in front of you: Make one edit, end to end, with the preview step How Hoard gates what an assistant can do ## If something doesn't work The two most common snags are the sign-in window getting blocked by a popup blocker, and the assistant saying it doesn't see any Hoard tools (usually a free plan or a connector that didn't finish connecting). Both are quick fixes in [common setup issues](/troubleshooting/assistant-connection). ## Where to go next * [What your assistant can do](/agent-capabilities) — the honest map of reads and changes * [AI assistant FAQ](/ai-assistant-faq) — is it safe, what does it cost, do I need my computer on * [Permissions and safety](/agent-permissions) — the preview → confirm → commit model * [Common setup issues](/troubleshooting/assistant-connection) — when the connection won't take