Access Guidance & Guardrails by selecting an agent from the Agents list and clicking the Guidance & Guardrails tab.
Primary configuration
Start by defining the fundamental regulatory context for your agent.Product type
Select the type of product your agent supports:| Product type | Description |
|---|---|
| Therapeutic | Pharmaceutical drugs and medicines |
| Diagnostic | Diagnostic tests and devices |
| Device | Medical devices |
Jurisdiction
Choose the regulatory jurisdiction that applies to your product:| Jurisdiction | Description |
|---|---|
| TGA (Australia) | Therapeutic Goods Administration — Australian regulatory framework |
| FDA (United States) | Food and Drug Administration — United States regulatory framework |
Primary audience
Confirm the intended audience for this agent:- Patient — General public and consumers
- HCP — Healthcare professionals
The primary audience is typically set when you create the agent, but you can confirm or adjust it here. This selection influences which compliance rules and guardrails are applied.
Therapeutic classification
The classification options depend on the jurisdiction you selected. Each classification carries specific regulatory implications that affect how your agent communicates.- TGA (Australia)
- FDA (United States)
For therapeutic products regulated by the TGA, select the appropriate schedule:
| Schedule | Name | Description |
|---|---|---|
| S2 | Pharmacy Medicine | Available from pharmacies without a prescription. Can be advertised to the general public with appropriate disclaimers. |
| S3 | Pharmacist Only Medicine | Available from pharmacies but must be handed over by a pharmacist. Advertising is permitted with restrictions. |
| S4 | Prescription Only Medicine | Requires a prescription from a qualified practitioner. |
Diagnostic test classification
If your product type is Diagnostic, additional classification fields appear.Test type
| Type | Full name | Description |
|---|---|---|
| IVD | In Vitro Diagnostic | Laboratory-based diagnostic test |
| LDT | Laboratory Developed Test | Test developed and used within a single laboratory |
| DTC | Direct-to-Consumer | Test marketed directly to consumers |
| POCT | Point-of-Care Test | Rapid test performed at or near the patient |
Test purpose
| Purpose | Description |
|---|---|
| Screening | Initial testing to identify potential conditions |
| Diagnostic | Confirmatory testing to establish a diagnosis |
| Monitoring | Ongoing testing to track disease progression or treatment response |
| Companion | Testing linked to a specific therapeutic product |
| Complementary | Testing that provides additional clinical information to inform treatment decisions |
Disease, indications, and terminology
This section defines the approved clinical language your agent may use.Approved disease references
List the diseases and conditions your agent is permitted to discuss. Only diseases listed here will be referenced in responses — the agent will avoid discussing conditions outside this list.Approved indications and claims
Define the specific indications and claims that have been approved for your product. These are the statements your agent is authorised to make — for example:- “Indicated for the treatment of moderate to severe plaque psoriasis in adults”
- “Reduces the frequency of migraine headaches in patients with episodic migraine”
Prohibited terms
Compass allows you to define terms and phrases that your agent must never use. Prohibited terms are organised into five categories:General
General
Common terms that could be misleading or non-compliant in a healthcare context.Default examples: “cure”, “miracle”, “breakthrough”, “guaranteed”, “100% effective”
Biological / Mechanism
Biological / Mechanism
Terms relating to biological processes or mechanisms that may overstate the product’s action.Default examples: “kills bacteria”, “destroys cancer cells”, “eliminates the virus”
Performance / Comparative
Performance / Comparative
Claims that compare the product to competitors or overstate its performance.Default examples: “best in class”, “superior”, “outperforms”, “number one”
Medical Advice
Medical Advice
Phrases that could be interpreted as providing direct medical advice.Default examples: “you should take”, “stop taking your medication”, “this will help you”
Custom
Custom
Add any additional terms specific to your product, organisation, or regulatory requirements.
Each category comes with sensible default terms. Review these defaults and add any additional terms specific to your product or regulatory environment.
Approved alternatives
For each prohibited term, you can map an approved alternative — a compliant phrase the agent should use instead.| Prohibited term | Approved alternative |
|---|---|
| ”cure" | "may help manage symptoms" |
| "miracle" | "clinically studied" |
| "guaranteed" | "has been shown in clinical trials" |
| "best in class" | "demonstrated efficacy in clinical studies” |
Off-label content policy
Configure how your agent responds when a conversation moves into off-label territory — that is, when a user asks about uses, dosages, or populations not included in the approved labelling.| Policy | Behaviour |
|---|---|
| Block | The agent refuses to respond to off-label queries entirely and informs the user that the topic is outside its approved scope. |
| Allow with Disclaimer | The agent provides a response but includes a clearly visible disclaimer stating that the information relates to off-label use. |
| Redirect | The agent does not answer the off-label query directly but redirects the user to an appropriate resource, such as a medical professional or a specific URL. |
| Escalate | The agent flags the conversation for review by your team and informs the user that their query has been escalated to a qualified representative. |
Best practices
- Start with stricter settings — It’s easier to loosen guardrails after review than to tighten them after a compliance issue
- Review default prohibited terms — The defaults are a solid starting point, but every product has unique regulatory considerations
- Align with your MLR process — Ensure your guardrail configuration has been reviewed by your Medical, Legal, and Regulatory (MLR) team
- Test thoroughly — Use the Batch Testing feature to verify your guardrails work as expected before publishing