Resources

Help & articles

Guides, how-tos and news for your catering business.

Help topics

Reports & Integrations

Connecting with an AI Agent

Give ChatGPT or another AI assistant read-only access to your Puree customers, items and orders, in plain English, and keep your access token safe.

You can give an AI agent read-only access to your Puree account, so that when you ask it a question about your business it fetches the real answer from your data instead of guessing at one. “Agent” and “assistant” mean the same thing here: ChatGPT, Claude and the like. This page is the plain-English version. If you or the person helping you wants the technical detail, that’s on The Developer API.

What It Can and Can’t Do

An assistant connected this way can look, but not touch. It can read three things: your customers, your menu items, and your quotes and orders. That’s enough to answer questions like:

  • Which quotes are expiring in the next fortnight, and who do I need to chase?
  • What did we sell most of last winter?
  • Which of our items are flagged as containing nuts?
  • List every accepted order for the last Friday of the month, with the delivery times.

It cannot write a quote, change a price, delete a customer or send an email. There is no setting that lets it. That’s worth knowing before you start, both because it sets expectations and because it means a mistake by the assistant can’t damage anything.

What You Need

Two things, and the second one is just a web address:

  1. An access token — a long password that identifies your account. You create it in Puree.
  2. The schema addresshttps://puree.app/api-docs/v1/swagger.yaml. Think of it as a menu: it tells the assistant which questions it’s allowed to ask and what the answers look like. It’s the same for everybody and it isn’t secret.

Step by Step

1. Ask us to switch API access on. It’s off by default while the API is in beta, and we enable it account by account. Email email@puree.app and we’ll turn it on for you.

2. Create your token. Go to Account Settings → Developer → Access Tokens and add one. You’ll be asked for three things:

  • A name. Something that says where it’s going, like “ChatGPT assistant”. If you later have several, this is how you tell them apart.
  • Scopes. Tick the areas this token may read: orders, customers, items. Tick only what the assistant actually needs.
  • An expiry. Optional. Leave it blank and the token lasts until you revoke it, or set a date and it stops working by itself. A date is the safer choice if you’re only trying something out.

The token appears once, on the screen straight after you create it. Copy it then. If you lose it you can’t look it up again, but you can always revoke it and make another.

3. Put the token into your assistant’s settings — not into the chat. More on this below, because it’s the part people most often get wrong.

4. Give it the schema address in whatever field your tool calls “schema”, “OpenAPI URL” or “import from URL”.

5. Ask it something. From here you talk normally. The assistant works out which lookup answers your question, sends your token along with the request, and reads the result back to you.

Will the Assistant Accept the Token?

Yes. Any tool built to connect to an outside service has a proper place to store a credential, separate from the conversation. In a ChatGPT custom GPT, for example, it’s under Actions → Authentication: choose API Key, set the type to Bearer, and paste the token there. It’s stored on your account, sent quietly with each request, and never shown in the chat.

Put the token in the credential field, never in a chat message.

Anything you type into a conversation is kept in that conversation’s history, can be read by anyone you share the chat with, and may be visible to other people on a shared workspace. A credential field is built for secrets; a chat box isn’t.

Assistants vary in what they support. Some connect through a newer method called MCP, which Puree doesn’t yet offer a ready-made connector for. If yours doesn’t have an obvious place for a schema address and an API key, that’s the likely reason, and it’s worth asking us before you spend an evening on it.

What About Zapier and Similar Tools?

Yes, with one caveat about how it has to be set up.

There’s no ready-made Puree app in Zapier’s directory yet, so you won’t find us by searching. What you use instead is Zapier’s own Webhooks step: add a GET request, put the address of what you want in the URL, and add a header called Authorization with the value Bearer followed by your token. That’s the same token and the same secrecy rules as everywhere else on this page. Zapier’s Webhooks step is a premium one, so it needs a paid Zapier plan. Make.com does the same job with its HTTP module, which is on its free tier.

The caveat is which way round it works. Puree doesn’t yet send a signal out when something happens, so nothing in Puree can start a Zap by itself. Instead you use Zapier’s Schedule trigger to go and look on a timetable. A Zap that runs at seven each morning, fetches the day’s orders and posts a summary into Slack works nicely. A Zap that fires the instant a customer accepts a quote isn’t possible yet.

And because the API is read-only, this is a one-way street: Zapier can pull information out of Puree, but it can’t create or change anything in your account.

Keeping the Token Secret

Treat the token exactly as you’d treat your Puree password. Anyone holding it can read your entire customer list and every quote you’ve written, from anywhere, without signing in as you.

  • Don’t email it, don’t put it in a shared document, and don’t paste it into a chat message.
  • Use a separate token for each tool. If one needs replacing you revoke just that one, and everything else keeps working.
  • If you share or publish a custom GPT, everyone who uses it is using your token and seeing your account’s data. Keep an assistant connected to your business data private to you and your team.
  • Check the Access Tokens screen now and then. It shows when each token was last used, so a token being used at a time you weren’t working is a signal worth acting on.
  • Revoke on the same screen the moment you’re suspicious, or when someone leaves, or when you stop using a tool. Revoking takes effect immediately. Nothing else about your account changes, and you can make a fresh token straight afterwards.

Because the access is read-only, the worst case if a token does get out is that someone reads your data. They can’t alter it or run up charges. That’s the reason we built it read-only first.

A Couple of Limits

Your token only ever sees your own account — there’s no arrangement under which an assistant could reach another caterer’s data, or another caterer’s assistant reach yours. And a token can make 60 requests a minute, which is plenty for asking questions and getting answers, but not enough to pull down your entire history in one go.

We’re Happy to Help

You don’t have to work this out alone. If you’d like a hand getting an assistant connected, or you’re not sure whether the thing you want to ask it is possible, email email@puree.app and we’ll walk you through it, or set it up with you on a call. We’d rather do that than have you give up on it.