Infrastructure
A policy grammar for writing rules, six governance APIs designed to be called independently, and a managed platform built on both. Primary application hosting: Frankfurt.
This page is about what the platform is made of and where it sits. How a single request travels through it is on /how-it-works and /architecture.
The layers
Each layer stands on its own. Together they make one conversation auditable from the first utterance to the receipt that closes it.
Layer 1
Public release in preparation
A YAML grammar for writing per-jurisdiction policy, and the thing the engine actually evaluates. It is written to be readable by a lawyer as well as by a machine, and it is headed for an open licence and a public repository.
Layer 2
Pre-dispatch validation, Article 50 attestation, the evidence vault, outbound eligibility, channel continuity and the multi-regulation bundle. Composable, and designed to be callable without the managed platform; no public API is served on api.audact.ai yet, apart from the published Merkle roots.
Layer 3
Controlled testing
Multi-tenant and white-label, for agencies that run governed voice on behalf of their own clients. Voice runs in controlled testing; chat is planned.
The building blocks
Every one of them has its own reference page with the call shape, the response fields and the questions people actually ask.
The mechanisms behind these APIs belong to a patent-pending governance stack (UK IPO).
Call shapes, response fields, and what each field is a record of.
Where we stand
Each line below says what it is, and just as importantly what it is not. Nothing here is an approval of Audact by anyone.
Public record
The AI Pact is a voluntary public commitment. The Commission states that the pledges are not legally binding, so signing confers no certification, no approval and no legal presumption that a system conforms to the AI Act. Acknowledgement of a consultation response is not an endorsement either.
The five positions we put to the EU AI Office, published in full.
What accrues
In controlled testing
A governed call does not only produce a receipt. It produces a record of the decision behind it: which rule was evaluated, what was allowed or stopped, in which jurisdiction, on which policy version. Production export of the receipts was retested on 26 August, measured on browser calls rather than over a telephone line. The record itself is kept per tenant — it is what an incident reconstruction is built from months later, when nobody remembers the call.
The engine is built so those records can be read back per jurisdiction and per vertical by the people who author the policy. That is a design property of the system, not a published benchmark and not a claim about accuracy.
Partners
Partner enquiries are welcome and are answered by a person.
The integration plan, and where each system stands.
The governance connectors, per system.
Direct lines
Press and partnership enquiries reach a founder, not a queue. If your question is about procurement, the Trust Center answers most of it before you write.
Press and analysts.
Consultancies, technology partners, integrations.
Sub-processors and hosting.
