Skip to main content

​
AI Commands

An AI Command is a reusable saved prompt in ChemCloud AI. It turns a task you do repeatedly — like a weekly delivery report — into a command you can trigger in any chat by typing its slug (e.g. /weekly-delivery-report). The best way to create a command is with the /create-command command, which is seeded into every buyer profile and walks you through building, testing and saving your own.

​
Commands use cases

Think of any repetitive task you do throughout the week:
  • Create a summary of incoming deliveries, to email to a customer or colleage
  • Check price history of a product, and compare to your most recent quotes
  • Review suppliers, for DIFOT, NCRs and price movements across key products
Commands are a way to perform these tasks quickly, in a standardised way. Importantly, you can share these across your entire team.

​
Triggering a command

  • Type the command’s slug in any chat, e.g. /weekly-delivery-report
  • Start a chat from the + interface in Chat
  • Find commands via global search

​
Creating a command

  • Type /create-command in any chat — this is the primary method. It interviews you about the task, drafts the command, test-runs it on your real data, and opens the New Command form pre-filled for you to save
  • Saved commands can also be viewed and updated directly from the commands interface

​
Anatomy of a command

Every command has four parts:

​
Structure of the instructions

Command instructions follow a fixed shape, in this order:
  1. Role and task — who the assistant acts as and the single outcome it produces
  2. Data sources — the specific ChemCloud records, reports or documents to use
  3. Steps — a numbered procedure, in the order you’d do it yourself
  4. Inputs — what you must supply at run time, and what happens if an input is missing (ask, or use a stated default)
  5. Output format — exactly what the command produces (table, summary, etc.)
  6. Constraints — rules the command always follows (see below)

​
Configuration options

Saved commands can be edited at any time from Saved Commands.

​
Examples

​
Creating a command with /create-command (primary method)

Run /create-command in any chat. It works in three phases:
1

Interview

It asks four questions, one at a time: what you’re trying to achieve, the steps you go through now, what inputs it will need, and how you want the output.
2

Test drive

It drafts the command (name, description, slug, full instructions) and runs the draft on a real example of your data so you can see actual output before saving.
3

Refine and save

It asks what to change, revises the draft, and opens the New Command form pre-filled so you can set visibility and save.
Nothing is saved until you’ve seen it working and confirmed.

​
Running a saved command

Type a saved command’s slug in chat and supply any required inputs:
The command runs its saved instructions against that customer’s open deliveries and returns the output in the saved format — for example, a RAG status table (Delivery ref, Product, Supplier, Promised date, Current ETA, Documents, Status) ordered Red → Amber → Green, with three bullets summarising what needs action this week.