Kyvvu

For legal and compliance

Sign off on agents you can actually inspect.

You're accountable for what the organisation's agents decide, but you can't read their code or predict their output. With Kyvvu you can read the rules they're held to, and see the record of what they did.

contract-agent · task 6a04kyvvu engine · in-process
  1. step.resourcePOSTtranslation/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

Accountable for decisions you can't see.

  1. 01

    Black boxes, real liability

    When an agent decides about a customer, the organisation is accountable for the outcome. 'The model did it' isn't an answer anyone accepts.

    accountability

  2. 02

    Policies nobody enforces

    Your AI policy, privacy rules and contractual commitments are written in documents. Agents don't read them.

    policy ≠ practice

  3. 03

    Reconstructing what happened

    When a decision is challenged, you need to show what the agent did, in what order, under which rule. Usually that means asking engineers to dig.

    disputes/audits/regulators

In practice

Where the law draws a line, the agent stops.

No decision on personal data without a lawful basis

An agent assesses a customer's request using their personal data. The rule: a person confirms consent or another lawful basis before the model runs.

assessment-agent · task 9d33kyvvu engine · in-process

Illustrative scenarios. Each one is decided by a single policy from the open policy library, as the engine evaluates it. References to laws and regulations are our reading of them, to show how a policy maps to an obligation. Not legal advice.

How Kyvvu helps

Rules you can read. A record you can show.

  1. 01

    Readable rules

    Each policy is a short, named rule in plain YAML, such as 'Erasure operations require an approval gate'. You can review them like any other control.

    policies as code

  2. 02

    A person where it matters

    A policy can require a person's approval before a decision is written, a deletion runs or a message goes out.

    human approval gates

  3. 03

    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

  4. 04

    The same answer every time

    Verdicts are deterministic: the same path gets the same verdict, so a decision can be replayed and explained.

    deterministic verdicts

Questions

Common questions.

01

Is this legal advice?

No. The compliance packs are our reading of the regulations, to show how a policy maps to an obligation. Your own legal assessment decides which rules apply.

02

Can we read and change the rules ourselves?

You can read them; they're short named rules. Changes go through your own repository, reviewed like code, so there's a history of who changed what and when.

03

What do we get for an audit or a dispute?

The trace of the task: every step in order, the verdict on each, and which rule decided, hash-chained when it reaches the platform.

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 board

The executive briefing kit

Ten slides and a one-pager on AI agent risk and accountability.

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