Assistant Settings
Detailed reference for Access Level, Conversation Starters, AI Model, Temperature, and Performance Metrics.
These settings apply to all three configuration types (Prompt, Structured, Nudge). You configure them when creating or editing any assistant.
Fields
Access Level
Controls who can use the assistant.
| Level | Who can access |
|---|---|
| Private | Only users you invite or authenticate with direct access |
| Restricted | Anyone with a valid cohort registration code |
| Public | Open to any authenticated user |
An assistant's access level cannot be more permissive than the course it belongs to.
| Course Access Level | Allowed Assistant Access Levels |
|---|---|
| Private | Private only |
| Restricted | Private or Restricted |
| Public | Private, Restricted, or Public |
The same rule applies when linking to multiple courses. Every linked course must be at least as restrictive as the assistant.
Practical implications:
- If your course is Private, your assistant must be Private.
- If you want a Public assistant, every course it draws from must also be Public.
- Restricted assistants can draw from Restricted or Public courses, but not Private ones.
Cohort integration: Assistants can be assigned to cohorts (groups of learners who share a registration code). In the assistant's detail view, use "Add to Cohort" and select from available cohorts. The same compatibility model applies:
| Cohort Access Level | Compatible Assistant Access Levels |
|---|---|
| Private | Private, Restricted, Public |
| Restricted | Restricted, Public |
| Public | Public only |
When a learner registers with a cohort code, they get access to all assistants assigned to that cohort (subject to access levels). This is how you deploy an assistant to a specific group without making it publicly available.
Conversation Starters
Between 1 and 5 starter prompts shown to learners when they first open the assistant. They reduce blank-page friction and set expectations for what the assistant can do. Each must be non-empty.
Good conversation starters are specific, actionable, and communicate the assistant's scope. They invite the learner to bring real work.
Good:
- "I'm working through a case and need help applying a framework. Where do I start?"
- "Can you walk me through how to score options in a weighted evaluation matrix?"
- "I have an assignment due tomorrow. Help me structure my analysis."
Weak:
- "Hello"
- "How can I help you?"
- "Ask me anything"
For Nudge assistants, conversation starters should set up the analytical task the path is designed to support:
- "I have a strategic decision to analyse. Let's work through it."
- "Walk me through the market entry framework for my case."
AI Model
The language model that powers the assistant. Defaults to Auto.
| Model | Provider | Supports Temperature | Supports Top-p |
|---|---|---|---|
| Gemini 2.5 Flash | Yes | Yes | |
| Gemini 2.5 Pro | Yes | Yes | |
| GPT-4o | OpenAI | Yes | Yes |
| GPT-4o Mini | OpenAI | Yes | Yes |
| GPT-4.1 | OpenAI | Yes | Yes |
| GPT-4.1 Mini | OpenAI | Yes | Yes |
| GPT-5 Mini | OpenAI | No | No |
| GPT-5 Nano | OpenAI | No | No |
| Claude 3.5 Haiku | Anthropic | Yes | Yes |
| Claude 3.7 Sonnet | Anthropic | Yes | Yes |
Temperature
Controls response creativity and variability. Range: 0.0 to 1.0.
- 0.0 to 0.3: Consistent, focused, deterministic. Best for analytical step-by-step work.
- 0.4 to 0.6: Balanced. Good default for most teaching assistants.
- 0.7 to 1.0: More creative, less predictable. Rarely appropriate for structured learning.
Default: 0.7
Top-p
Controls vocabulary diversity. Range: 0.0 to 1.0. Safe to leave at default unless you have a specific reason to adjust.
Default: 0.9
For models that do not support Temperature (GPT-5 Mini, GPT-5 Nano), both Temperature and Top-p are ignored regardless of what you enter.