Managing Assistants
Create, configure, and manage AI assistants for your teams.
Overview
Assistants are the core unit of Ajutant. Each assistant represents a specific role or capability — HR Assistant, Legal Assistant, Contract Reviewer, General Assistant — and is fully configured by your admin team.
Users never interact with raw AI models. They interact with assistants that your admins have purposefully designed.
Creating an Assistant
Navigate to Admin → Assistants → Create Assistant.
Required Configuration
Name — A clear, descriptive name that tells users what this assistant does. Examples: “HR Policy Assistant”, “Document Analyst”, “Contract Reviewer”.
Description — A short summary shown to users on the dashboard. Keep it to one or two sentences.
Model — Select which AI model powers this assistant. Models are assigned by tier:
| Tier | Use Case | Example Models |
|---|---|---|
| Reasoning | Complex analysis, multi-step logic | o3-mini, DeepSeek-R1 |
| General | Everyday tasks, chat, summarisation | GPT-4o |
| Economy | Simple queries, high-volume tasks | GPT-4o mini |
System Prompt — This defines the assistant’s behaviour, expertise, and constraints. It’s the most important configuration field. See Writing System Prompts below.
Optional Configuration
Temperature — Controls response creativity. Lower values (0.1–0.3) produce more consistent, deterministic responses. Higher values (0.7–1.0) produce more varied, creative output. Default: 0.3.
Knowledge Base — Attach document collections that the assistant can reference. See Knowledge Bases for details.
Team Access — Select which Azure AD groups can access this assistant. If no groups are selected, the assistant is available to all authenticated users.
Writing Effective System Prompts
The system prompt is the single most important factor in assistant quality. A well-written prompt produces consistently useful responses. A vague prompt produces inconsistent results.
Structure
A good system prompt covers four areas:
- Role — Who is the assistant? What is its expertise?
- Behaviour — How should it communicate? What tone?
- Constraints — What should it avoid? What are its boundaries?
- Format — How should it structure responses?
Example: HR Policy Assistant
You are an HR Policy Assistant for [Company Name]. Your role is to help
employees understand company policies, procedures, and benefits.
Behaviour:
- Be helpful, clear, and professional
- Reference specific policy sections when answering
- If a question falls outside HR policy, say so clearly
- Never provide legal advice — direct employees to the legal team
Constraints:
- Only answer questions related to company HR policies
- Do not speculate about policies you haven't been given
- If unsure, say "I'd recommend checking with your HR team directly"
- Never discuss individual employee cases or personal situations
Format:
- Keep answers concise — 2-3 paragraphs maximum
- Quote relevant policy sections when applicable
- End complex answers with "Would you like me to clarify anything?"
Editing and Disabling Assistants
Edit — Changes to an assistant take effect immediately for new conversations. Existing conversations continue using the configuration at the time they were started.
Disable — Disabling an assistant removes it from the user dashboard but preserves all conversation history. Disabled assistants can be re-enabled at any time.
Delete — Permanently removes the assistant and all associated configuration. Conversation history is retained but the assistant reference is removed.