You can access General Settings by selecting an agent from the Agents list and clicking the General tab.
Agent name
Give your agent a clear, descriptive name that helps your team identify it at a glance. The name is visible in the Compass dashboard and in widget configurations.- Maximum length: 255 characters
- Choose something meaningful — for example, “Aspirin HCP Detailer” or “Cardiology Patient Navigator”
AI model selection
Compass supports a range of AI models, grouped by speed and capability. Each model consumes a different number of credits per message.Fast models
Best for high-volume, straightforward interactions where speed is a priority.| Model | Credits per message | Notes |
|---|---|---|
| GPT-5 Mini | 0.5 | Recommended for most use cases |
| Claude Haiku 4.5 | 0.5 | Fast and capable alternative |
| GPT-5 Nano | 0.25 | Most cost-effective option |
Balanced models
A strong middle ground between speed and depth of reasoning.| Model | Credits per message | Notes |
|---|---|---|
| Claude Sonnet 4.5 | 1 | Excellent comprehension and nuance |
| GPT-5.1 | 1 | Strong general-purpose performance |
Reasoning models
Best for complex clinical queries that require deeper analysis and multi-step reasoning.| Model | Credits per message | Notes |
|---|---|---|
| GPT-5.2 | 1.5 | Advanced reasoning capabilities |
| Claude Opus 4.5 | 2 | Highest capability, best for complex queries |
Credit consumption is per message. Higher-tier models provide more thorough responses but use more credits. Monitor your usage on the Dashboard.
Agent instructions
The agent instructions determine the core behaviour of your agent. This is set automatically based on the agent type you selected during creation:- HCP Product Detailer — The agent is instructed to provide detailed clinical and product information to healthcare professionals using professional medical terminology.
- Patient Navigator — The agent is instructed to communicate in simple, accessible language, avoid medical advice, and follow strict compliance guidelines.
Product and therapy area
Depending on your agent type, you’ll configure one of the following:- HCP Product Detailer
Product
Select the product that this agent will detail. The agent uses this to focus its responses on the specific drug or therapeutic product, drawing from the knowledge base materials you’ve uploaded.Additional instructions
Use additional instructions to give your agent custom behavioural guidance beyond its default instructions. This is useful for fine-tuning how the agent responds in specific scenarios. To enable additional instructions:- Toggle the Additional Instructions switch to on
- Enter your custom instructions in the text field
- “Always recommend patients speak with their doctor before changing medication”
- “When discussing dosing, always reference the approved prescribing information”
- “If the user asks about pricing, direct them to the website”
Tone of voice
Set the conversational tone your agent uses when responding. The available options depend on the agent type.Response length
Control how much detail the agent provides in each response using the response length slider. There are four levels:| Level | Target length | Best for |
|---|---|---|
| Minimal | ~500 characters | Quick, focused answers to simple questions |
| Short | ~1,000 characters | Concise responses with essential detail |
| Long | ~2,000 characters | Thorough answers with supporting context |
| In-depth | ~4,000 characters | Comprehensive responses with full explanations |
These character targets apply to web-based widget responses. Actual response length may vary slightly depending on the complexity of the question and the AI model used.
Publishing your agent
Agents must be published before they can serve responses through widgets.- Draft agents
- Active agents
Draft agents show two buttons at the bottom of the settings page:
- Save — Saves your changes without publishing. The agent remains in Draft status.
- Publish — Saves your changes and sets the agent to Active status, making it available to connected widgets.
Danger zone
At the bottom of the General Settings page, the Danger Zone provides two irreversible actions:- Archive agent — Moves the agent to Archived status. It will be disconnected from all widgets and can no longer serve responses.
- Delete conversations — Permanently removes all conversation history associated with this agent.