Introduction

A key is a person, for a few hours

A Panlytics key lets a program or an AI agent call the API as the person who made the key. Everything it does is recorded under that person's name and marked as done with a key, and the person can see it in the app.

A key never reaches further than its maker. If your role covers two shops, a key you make covers the same two shops. If your role lets you see the figures but not export them, so does the key. When your role changes, the key's reach changes on its next call.

On top of that, a key is fenced. Whatever its maker's role holds, a key can never manage people or roles, read the audit log, change billing, security or the company's data, change what the agents may spend, or make another key. What an AI can never do lists the fence in full.

Short lived, by design

A key ends 1 to 5 hours after it is made. The company's CEO sets the period on the company's Developer page, and shortening it shortens every live key too. A key also ends at once when it is revoked, when developer mode is switched off, when its maker is removed from the company, or when its maker changes their password and signs out everywhere else. Keys and rotation covers every way a key ends.

A short key is a small loss when it leaks. It also means an agent never holds lasting access to your shops: when the work is done, the key runs out on its own.

What answers today

This is the first release of the developer API.

The API reference has a page for every endpoint. The agent prompt you copy from the app lists the same endpoints, so an agent is never told to call one that does not answer.

Where to go next

• Quickstart takes you from nothing to a first call in five minutes.
• Connect an AI agent shows where the prompt goes in Claude Code, Claude, ChatGPT and Cursor.
• How Panlytics is organised explains companies, shops, categories and suppliers, which every endpoint is built on.