A control layer that sits between your conversations and your models.
norse3 inspects inbound prompts and outbound responses, applies your policies and creates evidence — while people stay in control of decisions.
The control & evidence path
Illustrative signal flowUser
Person or system
AI / channel
Chatbot, copilot, agent
Bifröst Relay
Connect · normalise · route
Aegis Engine
Identify · decide
Allow / Warn / Block / Redact / Escalate
Configured action
Heimdall Console
Evidence · human review
The downstream model is not part of norse3. We inspect inbound prompts and outbound responses around it — we do not replace it.
Identify → decide → act → evidence → review.
Every high-risk interaction follows the same path, so behaviour is consistent and explainable.
Identify
Jailbreaks, PII exposure, manipulation, unauthorised advice, harmful content and policy breaches.
Decide
Evaluate against your configured policies, roles, jurisdiction and risk appetite.
Act
Allow, warn, block, redact or escalate — as your policy defines.
Evidence
Record the event, the policy applied, the action taken and the timestamp.
Review
Triage in a human-controlled console. Your firm remains accountable.
Relay, engine and console.
Each part has a clear job. Together they form one control and evidence path.
Bifröst Relay
Connects approved systems, normalises, encrypts and routes messages.
Aegis Engine
Detects, classifies, evaluates policy and applies the configured action.
Heimdall Console
Alert triage, incident review, policy management and reporting.
What happens to an interaction.
Three illustrative scenarios. Demonstrations only — not records of real customers.
A user submits a prompt engineered to override system instructions and extract restricted behaviour.
Aegis Engine matches the input against jailbreak and manipulation signatures and scores it high-risk.
The request is blocked before it reaches the downstream model. The user receives a safe fallback response.
No restricted behaviour is triggered. A high-severity event is written to Heimdall for review.
From discovery to ongoing monitoring.
A structured path that is validated with your teams at each stage. We do not promise deployment in minutes.
Discovery
Understand your conversations, systems, risks and obligations.
Policy mapping
Translate your policies into configurable, testable controls.
Integration
Connect approved channels through Bifröst Relay.
Testing
Validate detections and actions against agreed scenarios.
Launch
Go live with the controls and reviewers you have signed off.
Monitoring
Review evidence, refine policies and report on an ongoing basis.