A custom assistant is an AI that’s been configured for a specific task. It’s built on top of a model — GPT, Claude, or any other — but comes with structured instructions that shape how it behaves. Instead of starting from a blank slate, the assistant already knows what it’s looking at, what it should do, and how to present its findings.
Custom assistants are managed by administrators from the admin panel. If you
need an assistant for a workflow that doesn’t have one yet, or you want to
adjust how an existing assistant behaves, talk to your admin.
What makes them different
When you use a model directly, you’re responsible for all the context: explaining the task, specifying the format, defining the audience. A custom assistant handles that upfront. Its instructions are structured around five areas:
- Context — the background the AI needs, such as the type of documents it’s working with or the jurisdiction it should focus on
- Objective — what the assistant should accomplish when you ask it something
- Style — how it should communicate: formal, concise, detailed, or however your firm prefers
- Audience — who it’s writing for, whether that’s a partner, a client, or opposing counsel
- Response format — how to structure the output: bullet points, memo format, table, or something else
These fields are what separate a well-configured assistant from a generic prompt. They make the output consistent across users and conversations — so the firm’s contract review assistant produces the same style of analysis regardless of who’s using it.
Viewing assistant details
Before you use a custom assistant, you can see exactly what it’s designed to do. Hover over it in the assistant selector to see its description and which model it’s built on. Click View details for the full configuration, including all the instruction fields.
This is useful when you’re deciding whether an assistant fits your task, or when you want to understand why it’s giving you a particular kind of response. 
