Place the Kirtonic Engine in front of your AI providers. Messages are scored against your policy, risky outputs are queued for review, and each decision is recorded in an append-only audit log.
Define controls in plain English. The Engine categorises and enforces them across every request routed through it.
High-risk outputs are routed to designated reviewers. Each decision strengthens the classifier.
An append-only, tamper-evident record of signals, decisions and reviewer actions.
Reviewer decisions become labelled training data. Fine-tune a model on your own policy.
Type what should and shouldn't happen. The Engine categorises each rule for you.
One URL change. Every request flows through the Engine.
Reviewer decisions become labelled training data for your own classifier.
Keep personal data and regulated advice out of AI prompts, gate financial promotions against FCA rules, and hold a tamper-evident audit trail for every decision, the evidence a regulator expects.
Personal data, account identifiers and regulated advice are caught before a prompt reaches an AI provider.
Regulated advice and elevated-risk prompts route to a designated reviewer before anything is actioned, with every decision recorded.
Copy is gated against the FCA financial-promotions pack, kept current and dated, so what ships reflects the rules as they stand today.
An append-only, tamper-evident record of every signal, decision and reviewer action, exportable for audit.
A free browser extension checks prompts in real time and holds back secrets, personal data and payment details before they leave the browser.
Two deployment models, built around the data-sovereignty and operational needs of regulated organisations.
The Sandbox tier includes twenty classifications a month, so you can wire the Engine in and watch the audit log fill up.
