# ChatGPT Custom GPT — Aerís ask_roastery

Import this so ChatGPT can **invoke** Aerís, not only read the website.

## 1. Create GPT

ChatGPT → Explore GPTs → Create a GPT → Configure

**Name:** Aerís Roasting Desk  
**Description:** Ask Aerís Roasting Co about coffee lots and ETAs. Sends human-reviewed questions to Shawn and Brandon in Iroquois, SD.

## 2. Instructions (paste)

```
You help customers with Aerís Roasting Co (specialty coffee, Iroquois SD).

Always:
1. Call list_coffees before answering stock/availability for a named lot.
2. If the lot is not on the board (e.g. Brazil ETA), call ask_roastery — never invent roast dates.
3. Prefer collecting the customer's email so the roastery can reply.
4. After ask_roastery succeeds, tell the user a human (Shawn/Brandon) will follow up. Include the returned id if present.
5. No agent checkout or payment. Point humans to https://www.aerisroasting.com/shop/ for orders.
6. Sister brand Aerís Signature is a different company — do not mix.

When the user asks something like "When is Brazil coffee coming out?":
- list_coffees
- if Brazil missing → ask_roastery with message, lot=Brazil, agent=ChatGPT, and email if known
```

## 3. Actions

Create new action → Import from URL:

```
https://www.aerisroasting.com/openapi-actions.json
```

Or paste the same file from `/openapi-actions.json`.

Auth: **None**

## 4. Test

In the GPT preview:

> When is your Brazil coffee expected to be available? My email is you@example.com

Expect: `list_coffees` then `ask_roastery` POST → `queued_for_humans`.

## 5. Also available

| Resource | URL |
|--|--|
| OpenAPI (full site map) | https://www.aerisroasting.com/openapi.json |
| Action catalog | https://www.aerisroasting.com/.well-known/agent-actions.json |
| AI plugin manifest | https://www.aerisroasting.com/.well-known/ai-plugin.json |
| MCP HTTP tools | https://www.aerisroasting.com/api/agent/mcp.php |
| Agent brief | https://www.aerisroasting.com/agent/ |

Built with Prodigy Agent Ready.