Working with agents
Setting up a companion
A companion needs two things, its instructions and its memory. No setup files.
If you have used other AI tools, you may have a folder of soul.md, skill.md or agents.md files and wonder which ones Pragma wants. None of them. A companion in Pragma is set up by two things:
- Its instructions. Who it is, what it does, and how you want replies. They are in its profile: click its picture at the top of the chat, then Instructions. The engine reads them at the start of every conversation.
- Its memory. What it learns as you work: your name, your business, how you like things done. It keeps this itself, and you can read and correct it under Memory in the same profile.
That is the whole setup. Change either one at any time and the next message uses it.
Writing good instructions
Under the Instructions field, Help me write instructions asks five plain questions:
- What is its job?
- Who does it work for, and what is the business?
- What does good work look like?
- What must it never do?
- How should it sound?
The companion's own engine turns your answers into a draft. Read it, change anything, then Use these instructions. Nothing is saved until you do.
Short and specific beats long and general. "Reply to customer emails for my barbershop in two or three sentences, and never promise a discount" does more than a page about being helpful.
Skills are optional
A skill is a reusable how-to for one kind of task, like a report format or a review checklist. A companion works without any. Add one later under Teach a skill when you find yourself explaining the same procedure twice.
Installing an engine
To run companions on another engine, such as Codex, Claude Code or Qwen, ask any companion to install it, or open Settings, Engines. Pragma installs it and signs you in; you do not need to install it by hand. Then pick it in a companion's Model menu. See Engines and models.