Policy enforcement at runtime
Rules are enforced on every single turn, in the request path, not reviewed after the fact. A blocked action never reaches the model.
Regent
Regent: runtime governance for AI agents
Regent is a governance layer that sits between your people and the AI they use. Every turn passes through policy enforcement before it reaches the model, and every action lands in a tamper-evident audit chain after. If an auditor, a client, or your own compliance team asks "what did the AI actually do," Regent is the system that answers with proof instead of a shrug.
Book an intro callThe problem it solves
AI adoption stalls on one question: can we trust what it did? Logs can be edited. Screenshots prove nothing. Most AI governance today is a policy PDF nobody can enforce and a dashboard nobody checks. Regent replaces trust-me with show-me: mechanical enforcement at runtime, and records that break visibly if anyone tampers with them.
What Regent does
Tamper-evident audit chain
Every governed turn is written to a hash-linked chain. Edit one record and verification fails at exactly that line. This is the receipt for every AI action your organization takes.
Session governance and memory
Working context is preserved and governed across sessions, so the AI stays consistent and the organization, not the individual seat, owns the record.
Who it is for
Regent is built for organizations that answer to someone: MSPs answering to clients, contractors answering to agencies, regulated businesses answering to auditors. If "the AI did it" is not an acceptable incident report in your world, you are who we built this for.
How it fits
Regent works with the AI subscriptions you already have. It does not replace your models; it governs them. Deployment is a lightweight install alongside your existing tools, currently in alpha with pilot organizations.
Frequently asked questions
What is runtime AI governance?
Runtime AI governance means enforcing rules while the AI is working, on every request, instead of auditing after something has gone wrong. It is the difference between a guardrail and an incident report.
What is a tamper-evident audit trail for AI?
A record of AI activity where every entry is cryptographically linked to the one before it. If any entry is altered, verification fails and shows exactly where. Ordinary logs cannot make that promise.
Does Regent work with ChatGPT and Claude?
Yes. Regent governs the connection between your users and major AI providers, including OpenAI and Anthropic models, using your existing subscriptions.
Is Regent available today?
Regent is in alpha with pilot organizations. Book an intro call to talk about a pilot.