Guardrails & protections
The guardrails are what set Fred free.
Fred combines your business content with configured guardrails for sensitive topics. Set the boundaries, test the experience, and give visitors a clear path to your team.
A safer Fred is a faster Fred.
Clear boundaries help you decide what Fred should answer and what belongs with your team. Configure checks around those decisions, test realistic questions, and review the conversations that need attention.
Four layers of protection, stacked.
Build your configuration from baseline, regional, industry, and custom rules. Confirm which layers are active for your business and test the behavior before launch.
Baseline topics and boundaries
Baseline rules address configured patterns involving crisis language, violence, child safety, fraud, abuse, and sensitive information. Detection has limits. Review the active rules and escalation settings for your deployment.
The rules that change by country
Regional settings help tailor resource messages and privacy controls to where you operate. Confirm the supported region and local resources during setup. See the regional rules →
The pack for your trade
A dentist, a roofer, and a law firm need different conversation boundaries. Review the industry pack for your field and configure the topics reserved for your team. See it by industry →
Anything you want to add
The pack is the floor, not the ceiling. You can add unlimited rules of your own in plain language, block, warn, or quietly log on any phrase you choose. No developer required.
Fred evaluates configured rules in code outside the AI model. These checks target recognized patterns; they do not guarantee that every unsafe request or response will be detected.
Start with baseline guardrails.
Baseline rules provide a starting point for sensitive topics. Confirm the active configuration and test it for your business before launch. Automated checks can miss context or produce false alarms.
Someone in crisis
Configured crisis patterns are designed to interrupt the sales flow, surface crisis resources, and support a configured team escalation. Fred may miss crisis language and is not an emergency service. A notification does not guarantee a person has received or acted on it.
A threat of violence
Refusal rules target recognized threats and requests for harmful instructions. Detected matches can trigger a configured response or request for escalation; review the coverage and delivery workflow.
Anything involving a child
Child-safety rules target configured patterns and can request team attention. Detection and notification do not replace your safeguarding procedures or emergency services.
Fraud, hacking, or anything illegal
Refusal rules are designed to decline recognized requests for fraud, account intrusion, and other prohibited activity, with an appropriate next step where configured.
Hate, harassment, or explicit content
Configured refusal rules target hateful, harassing, and explicit content. Review their coverage and monitor conversations; automated checks cannot guarantee every harmful response will be prevented.
Sensitive personal information
Fred’s intended behavior is to direct sensitive financial and identity information to an appropriate secure channel. Visitors can still submit sensitive data. Redaction depends on the configured patterns and settings and can miss information.
Clear AI identification
Fred tells people they are talking with an AI assistant and offers a human for anything important. Honesty here is now the law in a growing number of places.
The three we are most serious about.
Three situations can quietly turn into a headline, a lawsuit, or worse. Here is why each one matters and exactly what Fred does about it.
A life comes before a lead
People in real distress reach out to whoever is there, even a furnace-repair chat at 2am. The wrong response is unthinkable.
- Configured crisis patterns are designed to interrupt the sales flow when detected
- Configured resource messages should match the supported region
- A configured escalation can request team follow-up; receipt and response depend on the receiving team
Never the tool, never the coach
An assistant that engages with a threat, even to argue, is a liability. Fred treats these as hard stops.
- Refusal rules target detected threats to people
- Rules are designed to decline instructions for weapons or attacks
- Configure escalation destinations and check how failed notifications are handled
Support accessible conversations
Customers who are blind, deaf, or use assistive tech have a legal right to equal access, and inaccessible sites draw real lawsuits.
- Fred notices when someone needs an accessible format
- It offers plain language, larger text, or another channel
- It connects the visitor to a person who can accommodate them
See how the crisis, accessibility, and stay-in-your-lane protections work →
Different country, different rules.
Resource availability and privacy requirements vary by location. Confirm the region, resource messages, and handling of personal information for your deployment. Regional settings do not establish legal compliance.
Set your boundaries. Review how they work.
Talk to Fred.
Get a Fred that knows where the line is.
Choose a plan, configure Fred for your business, and test before launch. Or ask Fred a question in the corner of this page.