Kyvvu

AI agent governance

Your AI policy, enforced on every agent.

Agents are being built in every team, on every framework. Kyvvu gives you one register of all of them and one set of rules that is enforced on each agent at runtime.

hr-assistant · task 0c3ekyvvu engine · in-process
  1. step.messagePOSTsmtp/recruiter-externalblock
Blocked: outbound send after personal data was read in this taskTry it: click the highlighted step to take it off the path

The problem

Most AI policies stay in a document.

Most organisations have an AI policy. Very few can show it's applied to the agents actually running.

  1. 01

    Nobody has the full list

    Agents start as experiments in one team and end up in production. Which ones run today, who owns them and what can they touch?

    shadow agents

  2. 02

    Every team builds its own controls

    One team uses LangGraph, another CrewAI, a third Claude Code. Each solves safety differently, if at all, and none of it adds up to one policy.

    fragmented controls

  3. 03

    Rules that stay on paper

    "A person approves anything destructive" is easy to write down. Proving that it happens on every run of every agent is another matter.

    policy ≠ practice

In practice

Your rules, applied to what agents actually do.

Three examples of a rule deciding a real step.

"A person approves anything destructive"

An operations agent tidies up old records. Your policy says deletions need a person's approval. The agent tries without it first.

ops-agent · task 9a12kyvvu engine · in-process

Illustrative scenarios. Each one is decided by a single policy from the open policy library, as the engine evaluates it.

How Kyvvu helps

One register and one policy for every agent.

  1. 01

    Every agent on the register

    Each agent registers with its purpose, owner, risk classification, the tools it may use and its environment. Registering is free.

    agent registry

  2. 02

    One policy, every agent

    The same rules for plain Python, LangChain, LangGraph, CrewAI and Claude Code agents, in the cloud, on-prem or air-gapped.

    Python/LangChain/LangGraph/CrewAI/Claude Code/any language

  3. 03

    Rules people can read

    Policies are short YAML files in your own git, reviewed and versioned like code. Every verdict in the trace names the rule that decided.

    policies as code

  4. 04

    Incidents your team can follow up

    Every warn or block opens an incident with the policy, its severity, the step and the agent's risk level.

    incidents

  5. 05

    Nothing breaks on day one

    Only critical policies block; high, medium and low ones warn. Start a new rule as a warning and make it critical once you trust it.

    shadow/suggest/enforce

  6. 06

    A record of every step

    Each task is recorded as an ordered trace of its steps with the verdict on each, hash-chained when it reaches the platform.

    logs/traces

Questions

Common questions.

01

Do we have to govern every agent?

No. Register every agent for free; you decide which production agents to govern, on top of a platform fee: €150 per agent a month on Team, €100 on Enterprise & Public. Test and staging agents are free.

02

Can a policy apply only to high-risk agents?

Yes. A policy can apply to one risk class and above, or to every agent. Each agent's risk classification comes from its registration.

03

How fast does a policy change reach running agents?

Within the policy refresh window, five minutes by default, with no redeploy.

Get started

Every agent has to ASK before it acts.

Start with one agent

See it on your own agents

A 30-minute demo with someone who builds Kyvvu: your agent, your rules, and the policy that decides each step.

For your executive committee

The executive briefing kit

Ten slides and a one-pager on AI agent risk, ready for your board or executive committee.

Want to see other teams and sectors? All use cases.