Skip to main content
Compass is made up of a set of interconnected concepts that work together to help you build, deploy, and monitor compliant AI agents. This page explains each one so you can navigate the platform with confidence.
New to Compass? We recommend reading through these concepts before diving into the detailed guides. It’ll make everything else much easier to follow.

​
How it all fits together

At a high level, the Compass workflow follows five steps:
1

Create an Agent

Choose your agent type and configure its AI model, tone, and response behaviour.
2

Add Knowledge

Upload documents and URLs to build your agent’s knowledge base.
3

Configure Guardrails

Set regulatory compliance rules, prohibited terms, and off-label policies.
4

Deploy a Widget

Create an embeddable chat widget, customise its appearance, and go live.
5

Monitor & Refine

Track conversations, review adverse events, and improve your agent over time.

​
Core concepts

Agents are AI-powered conversational assistants — the heart of the Compass platform. Each agent is configured to serve a specific product and audience.Compass supports two agent types:
  • HCP Product Detailer — Designed for healthcare professional audiences. Delivers detailed, clinical information grounded in approved materials. Ideal for product education and medical enquiries.
  • Patient Navigator — Designed for patient and consumer audiences. Uses accessible, supportive language to provide health and product information in a way that’s easy to understand.
Each agent has its own:
  • AI model — The large language model powering the agent’s responses (e.g. GPT-5 Mini).
  • Tone of voice — How the agent communicates (professional, empathetic, conversational, etc.).
  • Response length — How detailed or concise the agent’s replies are.
  • Compliance settings — Jurisdiction, guardrails, and regulatory configuration.
You can create multiple agents for different products, markets, or audiences — all managed from a single dashboard.
Sources are the documents and web pages that form your agent’s internal knowledge. When a user asks a question, the agent searches its knowledge base and retrieves the most relevant content to generate a grounded response.How it works:
  1. You upload PDFs or add URLs to the Knowledge Base.
  2. Compass automatically processes each document — splitting it into smaller chunks, generating embeddings, and indexing them.
  3. When a user asks a question, the agent uses retrieval-augmented generation (RAG) to find and reference the most relevant chunks.
This approach means your agent’s responses are grounded in your approved materials rather than general AI knowledge, significantly reducing the risk of hallucination or off-label content.
Sources are internal to the agent. End-users never see the raw source documents — they only receive the agent’s generated responses. If you want users to access files directly, use Resources instead.
Resources are downloadable files and links that are surfaced directly to end-users through the widget’s Resources tab.Unlike Sources (which power the agent’s responses behind the scenes), Resources are visible, browsable content that users can access for reference — things like product brochures, consumer medicine information leaflets, or links to external websites.Use Resources when you want to give users something they can take away or refer to alongside the conversation.
Guardrails are the regulatory compliance settings that control what your agent can and cannot say. This is one of the most important configurations in Compass.Key settings include:
  • Jurisdiction — The regulatory framework your agent must comply with (e.g. TGA for Australia, FDA for the United States).
  • Product classification — Whether the product is a prescription medicine, over-the-counter medicine, medical device, supplement, or other type.
  • Prohibited terms — Specific words or phrases the agent must never use in its responses.
  • Approved claims — Statements the agent is permitted to make about the product.
  • Off-label content policies — Rules for how the agent responds to questions about unapproved uses or indications.
Properly configured guardrails are essential for regulatory compliance. Always review and test your guardrail settings before deploying an agent to a live environment.
Topics allow you to track and categorise what users are asking about. Each topic is defined by a set of keywords — when those keywords appear in a conversation, the topic is automatically detected and logged.This gives you insight into user interests and frequently discussed themes, helping you:
  • Identify trending questions or concerns
  • Spot gaps in your agent’s knowledge base
  • Report on conversation themes to stakeholders
For example, you might create a topic called “Side Effects” with keywords like adverse, reaction, side effect, and symptom to track how often users ask about safety.
The Adverse Events feature is Compass’s built-in pharmacovigilance system. It uses AI to automatically detect potential adverse drug reactions mentioned in conversations.When a user describes symptoms, side effects, or negative experiences that may be related to a product, the system flags the conversation for review. You can configure the sensitivity level to control how aggressively the system flags potential events.Flagged events are surfaced in the Adverse Events dashboard, where your team can review, assess, and take appropriate action in line with your pharmacovigilance obligations.
Adverse event detection is a support tool — it does not replace your organisation’s pharmacovigilance processes. Always follow your established procedures for reporting and assessment.
Widgets are the embeddable chat interfaces that bring your agent to life on external websites. A widget is the user-facing component — it’s what your audience sees and interacts with.Each widget is linked to an agent and can be fully customised:
  • Widget type — Popup, fullscreen, or slider layout.
  • Theming — Brand colours, logos, and custom styling.
  • Tab configuration — Choose which tabs are visible (Chat, Resources, etc.).
  • Domain restrictions — Limit which websites can display the widget.
  • Status — Toggle between Active and Inactive without removing the embed code.
To deploy a widget, simply copy the embed snippet and paste it into your website’s HTML.
Initiatives are country-specific campaigns that group related deployments and widgets together. They provide an organisational layer that helps you manage multi-market rollouts.For example, if you’re launching a product in both Australia and New Zealand, you might create separate initiatives for each country — each with its own widgets, compliance settings, and reporting.
Batch Tests are Compass’s quality assurance system for validating agent responses before deployment. You define a set of test questions and expected behaviours, then run them against your agent to check that it responds correctly.This is especially useful for:
  • Verifying compliance after updating guardrails
  • Checking response quality after adding new knowledge
  • Regression testing before going live with changes
Run batch tests regularly — especially after making changes to your agent’s knowledge base or guardrails — to catch issues before they reach your users.
The Audit Log is a comprehensive activity trail that records every significant action taken across the platform. This includes agent creation, configuration changes, widget deployments, user logins, and more.The audit log supports your governance and compliance requirements by providing a clear, searchable record of who did what and when.
HCP Verification is a verification gate that ensures only qualified healthcare professionals can access restricted content. When enabled on a widget, users must verify their identity through AHPRA (the Australian Health Practitioner Regulation Agency) before they can interact with the agent.This is essential for HCP Product Detailer agents that surface clinical or restricted product information that should not be accessible to the general public.

​
What’s next?

Now that you understand the building blocks, start putting them into practice:

Quick Start

Follow the step-by-step guide to create and deploy your first agent.

Creating an Agent

Learn how to set up and configure an agent in detail.

Knowledge Base

Add documents and URLs to power your agent’s responses.

Deploying a Widget

Create and embed a chat widget on your website.