Skip to content
Service status

Type to search

7 articles · 4 min read

Create an assistant

Ask your personal assistant to create, set up and share an assistant, and what it checks before it changes anything.

An assistant is an AI set up for one job and shared with the people in a Space: its own instructions, Skills, knowledge and model, so everyone who uses it gets the same behaviour without writing the same prompt each time. Your personal assistant is for you alone; an assistant you create lives in a Space and is built for a team.

The quickest way to create one is to ask your personal assistant. Describe what you need in the chat, and it creates the assistant, writes its instructions, adds Skills and shares it, asking for your approval before each change.

You need the Creator or Admin organization role. On top of that, each change goes through the same permission checks as when you make it by hand:

  • Creating or editing an assistant needs the Space Editor or Space Admin role in the Space it belongs to. A Creator who is only a Viewer in a Space can’t add assistants there.
  • Creating a Space needs the Creator or Admin organization role.
  • Adding people to a Space or changing its security classification needs the Space Admin role in that Space.

If you ask for something your role doesn’t allow, the change is refused and the assistant tells you why. See Roles in Intric for what each role covers.

Open your personal assistant and describe the assistant you want: what it should do, who it is for, and which Space it belongs in. For example:

The personal assistant works out the details, then asks you to approve each change. When it has created the assistant, the result in the chat links to it, so you can open it and carry on in its settings.

These tools are only available in your personal assistant’s chat. Assistants in a Space can’t create or change other assistants or Spaces.

For assistants:

  • create an assistant in a Space you choose, with a name, description, instructions and model;
  • change an existing assistant’s name, description, instructions or model;
  • add a Skill, either one it writes from what you describe or a SKILL.md file you upload in the conversation;
  • read an existing Skill and change its title, description or instructions. A Skill uploaded as a bundle can only be renamed;
  • share an assistant with a colleague as a collaborator. The colleague must already be a member of the assistant’s Space, so the assistant adds them to the Space first if needed.

For Spaces:

  • create a Space, which makes you its Space Admin;
  • add a colleague to a Space as Viewer, Editor or Admin;
  • change a Space’s security classification.

Looking things up happens without asking: which Spaces and assistants you have, which models you may use, which security classifications exist, and who your colleagues are. Every change after that waits for your approval, one change at a time.

The approval names exactly what will happen: the colleague, the assistant, the Space and the model involved. Nothing changes until you approve it, and you can decline.

When you haven’t said which Space, model or classification to use and more than one would fit, the assistant asks you rather than picking one.

What to finish in the assistant’s settings

Section titled “What to finish in the assistant’s settings”

The personal assistant covers the basics: name, description, instructions, model, Skills and sharing. For knowledge, attachments, tools, web search, the transcription and image models, or publishing, open the assistant and finish in its settings. See Instructions, Skills, Knowledge and AI setup.

To build an assistant yourself from the start, see Create an assistant manually.

A new assistant doesn’t take anything from your personal assistant. Its model and classification come from the Space it is created in, exactly as when you create it by hand.

  • Classification. The assistant inherits its Space’s security classification. That classification decides which models, knowledge and tools the assistant is allowed to use.
  • Model. If you don’t name a model, the assistant gets your organization’s default model, as long as it is allowed in the Space’s classification. If it isn’t, Intric picks another model that is. If you name one, it has to be a model your organization has enabled and that the Space’s classification allows. Otherwise the change is refused.
  • A new Space. If your organization uses security classifications, every new Space needs one, and the assistant asks you which. Assistants you create in that Space then follow it.

Raising a Space’s classification can stop assistants in it from working, for example when their model isn’t allowed at the new level. Before it asks for your approval, the assistant shows which assistants would stop working and which of the tools they use would no longer be allowed. The approval lists those assistants again, and the change is refused if the list no longer matches what would actually happen.