Deal.ai

View saved version →

---
name: check-a-deal
description: "Check whether a price is a good deal: Deal.ai finds the exact item at every big US store right now. Use when the user shares a store link, asks whether a price is good, or asks if a sale or 'was' price is real."
---

# Check a deal

## How to run it

1. Call `check_deal` with the product: a link to it at any store, or its name (with model and size if the user gave them). Pass `price` when the user says what they were offered.
2. It answers at once with a `check_id`. Tell the user it's checking (it takes 20 to 70 seconds), then call `get_deal_check` with the `check_id`. Call it again every 10 to 15 seconds while its status is `running`.
3. When it's done, give the verdict sentence first (cheaper elsewhere, the lowest price, or the only store), then the stores' prices with their links, and whether the store's "was" price holds up when the answer has one.
4. End with the result page link: it's shareable.

## Rules

- A check that didn't finish says so: tell the user and offer to check again.
- Don't guess a verdict before `get_deal_check` returns one.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: compare-prices
description: "Compare one product's price at every store Deal.ai tracks, with its price history. Use when the user asks how much something costs, where it's cheapest, or whether its price has been lower."
---

# Compare prices for a product

## How to run it

1. If the user gave a Deal.ai product link or a product_id, use it. Otherwise call `search_products` with the product's name (kind `product`) and pick the result that is the exact product (model, size, generation). If several fit, ask which one in a short question, listing them.
2. Call `get_product_prices` with its `product_id`.
3. Reply with the best store price first (the one marked best), then the next few stores, each with when it was seen and its store link. Mention marketplace sellers only as such. Add the lowest price Deal.ai has tracked when it's useful ("it was $179 on Black Friday").
4. End with the product's Deal.ai page, and keep any store-link notes the answer ends with.

For specs, a description or ratings, call `get_product_details`. To weigh two to four products against each other, call `compare_products`.

## Rules

- Quote prices exactly as returned, with the day they were seen.
- Don't call a price the best if the tool marked it as a marketplace seller's.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: deals
description: "Find live deals on Deal.ai by search, store, category, discount or price. Use when the user asks for deals, sales or discounts on something, at a store, or in a category."
---

# Find live deals

## How to run it

1. Call `get_deals` with what the user asked: `query` for a kind of product ("air fryer"), `store` for a store ("Best Buy"), `category`, `min_discount_percent` or `max_price`, and a `limit` that suits the reply (5 is a good default).
2. List the deals: title, price, how much under the earlier price and what that earlier price is, the store with its link, and when it was checked. Keep any store-link notes at the end.
3. Offer the list page the answer links for more.

## Rules

- Say when each deal was checked; deals end and prices change.
- If nothing matches, say so and suggest a broader search.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: find-what-to-buy
description: "Ask Deal.ai's shopping agent what to buy: it researches the request and recommends picks with each one's best store price. Use for 'best X under $Y', 'what should I get for...', and follow-ups that narrow the picks."
---

# Find what to buy

## How to run it

1. Call `ask_deal_ai` with the user's request in their own words, including budget, must-haves and who it's for. Don't rewrite it into keywords.
2. It waits about 20 seconds. If the answer isn't ready, it returns `running` with a `search_id`: tell the user Deal.ai is researching (answers take about 1 to 3 minutes) and call `get_deal_ai_answer` with the `search_id` every 15 to 20 seconds until it's done.
3. Give the agent's recommendation in a few lines, then each pick: name, why it was picked, its best store price with when it was seen, and its store link. Keep any store-link notes at the end.
4. For a follow-up ("cheaper ones?", "only Sony"), call `ask_deal_ai` again with the same `search_id` and the follow-up in the user's words.

## Rules

- If the agent asks a question back, ask the user the same question, then pass their answer as a follow-up.
- Don't add picks or prices the agent didn't return.
- One question at a time in a conversation: wait for the last answer before a follow-up.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: get-started
description: "Start here with Deal.ai: what it can do in ChatGPT and which tool fits which question. Use when the user first connects Deal.ai, asks what it can do, or asks a shopping question and no other Deal.ai skill fits."
---

# Get started with Deal.ai

Deal.ai compares prices for the same product across US stores (Best Buy, Walmart, Target, Amazon, brand stores and more), tracks each product's price over time, checks whether a deal is real, and has a shopping agent that researches what to buy. The user's own Deal.ai account holds their wish list, which e-mails them when a watched product's price drops.

When the user asks what Deal.ai can do, answer in a few lines from the list above and suggest one or two things to try. Don't call a tool just to describe Deal.ai.

## Which tool for which question

- "What should I buy?", "best X under $Y": the find-what-to-buy skill (`ask_deal_ai`).
- "How much is X?", "where is X cheapest?": the compare-prices skill (`search_products`, `get_product_prices`).
- "Is this a good deal?", a store link with a price: the check-a-deal skill (`check_deal`).
- "Any deals on X?", "today's Best Buy deals": the deals skill (`get_deals`).
- "Tell me when it drops", "what am I watching?": the watch-prices skill.
- Details or specs of one product: `get_product_details`. Two to four products side by side: `compare_products`.

## Rules for every answer

- Prices are the stores' listed prices at the time shown, before tax. Say when a price was seen; prices change.
- Link the Deal.ai page each answer gives, and use a price's `buy_url` when the user wants to buy. Nothing is bought in ChatGPT: the user buys on the store's site.
- A price marked as a marketplace seller's (eBay, a third-party seller) is labelled as such and is never "the best price".
- When an answer ends with store-link notes (commission, Klarna, Amazon), keep them in your reply.
- Never state a price, store or verdict that a Deal.ai tool didn't return.

## When not to use Deal.ai

Questions with no shopping in them, arithmetic, and general product knowledge you can answer yourself. Deal.ai never buys anything, pays, or changes an order.

Referenced files: 1 — file references are metadata, not package contents.

View saved version →

---
name: watch-prices
description: "Watch products for price drops with the user's Deal.ai wish list, which e-mails them when a price drops. Use when the user asks to be told when something gets cheaper, asks what they're watching, or wants to stop watching something."
---

# Watch prices

## How to run it

- **Watch:** find the product with `search_products` if you don't have its `product_id`, then call `watch_price`. Tell the user it's saved, its best price now, and whether price-drop e-mails are on for their account. If they're off, say so and give the link the answer includes to turn them on; never claim an e-mail will come when they're off.
- **What am I watching:** call `list_watched` and list each product with its price now.
- **Stop:** call `stop_watching` with the product. Confirm what was removed.

## Rules

- Only watch or stop watching when the user asks for it.
- If the wish list is full, tell the user and offer to remove something first.

Referenced files: 1 — file references are metadata, not package contents.