Assistants
How AI assistants work for end users.
What Users See
When users log in to Ajutant, they see a dashboard of available assistants — not a list of AI models. Each assistant has a name, a description, and a clear purpose. Users pick the assistant that matches their task and start a conversation.
This design is deliberate. Research and real-world deployments consistently show that giving non-technical users a choice of AI models leads to confusion, misuse, and low adoption. Assistants remove that friction entirely.
Starting a Conversation
- Click on an assistant from the dashboard
- Type your message in the chat interface
- The assistant responds using its pre-configured model and behaviour
Conversations are persistent. You can close the browser and return later to continue where you left off. All conversations are stored in your tenant’s PostgreSQL database.
How Assistants Work Behind the Scenes
Each assistant is a configuration layer on top of a model:
| Component | What It Controls |
|---|---|
| System prompt | The assistant’s personality, expertise, and behavioural constraints |
| Model | Which AI model powers the assistant (e.g., GPT-4o, GPT-4o mini) |
| Temperature | How creative or deterministic the responses are (0.0–1.0) |
| Knowledge base | Which documents the assistant can reference (if any) |
| Team access | Which Azure AD groups can see and use this assistant |
Users don’t see or control any of these settings. That’s the admin’s job.
Conversation History
All conversations are automatically saved and searchable. Users can:
- View past conversations in the sidebar
- Continue any previous conversation
- Start new conversations at any time
Conversations belong to the user who created them. Team separation ensures that users in one group cannot see conversations from another group.