The Audact platform
Control decides, before it runs, whether a governed AI action stays inside what your organisation allows. Proof is what that decision leaves behind. Voice is the only one you can buy today.
Audact infrastructure
Two layers, and together they are what Audact is. Control is the mandate and authority layer: the ceiling, the scope, the approvers, and the refusal in front of a governed action. Proof is what that action leaves behind, one receipt at a time.
The organisation defines the mandate; Audact enforces it. We do not advise how much authority an agent should get: you set the principal, the ceilings, the scope, the approvers and the risk appetite.
In developmentInternal only
Decide what a governed AI action may cause on the company's behalf before it runs: the ceiling, the scope, who may approve above it, and what happens when the ceiling is reached. The architecture is described and the runtime is being built; there is no sale price, and there will not be one until the delivery gate is confirmed in writing.
See the evidence a governed decision leaves behind. Three fields on the receipt can be recomputed by anyone today, without a key and without an account: the input hash, the output hash and the link to the previous receipt. A fourth, inclusion in the Merkle root, is checked against the published root with the inclusion proof from the call's Article 50 report. Production export of a set of receipts worked when it was retested on 26 August 2026 after the deploy (measured on browser calls rather than over a telephone line), and there is no public verifier yet.
The failure contract differs per action class. Fail-closed is right for an action that commits money or makes a promise; reading a record or booking an appointment does not carry the same consequence. Which class falls back how is agreed per customer and written into the contract.
Built on Audact
In a conversation the commitment is made before anyone can review it. That is why voice is the first workload we put on the two layers above, and why it is the one thing on this page you can buy today. It is called Governed Voice AI.
In controlled testing
Inbound and outbound AI calls, with the Article 50 disclosure delivered at the boundary where the disclosure gate is enabled for that jurisdiction. Netherlands and Flanders (BE): intake open. Germany and Austria: intake is a conversation and not a start date.
In controlled testing
Eligibility, consent and quiet-hour checks run before a single number is dialled. The call goes out only if it clears.
Planned
The same disclosure duty and the same evidence chain carried into text, under one set of governance rules.
Netherlands and Flanders (BE): intake open. Germany and Austria: intake is a conversation and not a start date.
The engine
The same three components run under every channel.
In controlled testing
Where the disclosure gate is enabled for that jurisdiction, the disclosure is delivered at the boundary before the caller hears or reads a word.
A tamper-evident compliance receipt per interaction. A fourth field, inclusion in the Merkle root, is checked against the published root with the inclusion proof from the call's Article 50 report. The other three — input hash, output hash and the link to the previous receipt — can be recomputed by anyone today, without a key and without an account.
In controlled testing
Per-jurisdiction policy bundles and your brand rules check every candidate response in real time, before it reaches the customer.
Policy differs per country because the national layer on top of Article 50 differs. The German pack is active; the Dutch layer is the next one being built.
Named parts
These four have their own pages because each of them answers a question a buyer asks in those words. They are parts of Control or Proof, or items on the roadmap. None of them is sold on its own, none of them carries its own price, and the status below each one is the status it has today.
Not available yet
Part of Proof: a compliance receipt for every governed interaction. Three of its values recompute in your own browser, with the browser's own SHA-256. No key, no account, and nothing is sent to us.
Manual pack on request · generator: Planned
Part of Proof: the evidence bundle an auditor actually asks for (eight artefacts, among them a FRIA where the customer's deployment falls within Article 27, Article 50 templates and an AI literacy log), assembled by hand on request from your own receipts rather than from a Word template.
In the platform, not a separate product · In controlled testing
Part of Control: every consent event with its provenance, scope, expiry and revocation, linked to the interaction it governed, exportable on demand.
SLOs: In controlled testing · not a separate product
Across both layers: four SLO targets for regulated AI, with alerts that reach your incident channel before a compliance gap turns into a regulatory event.
Capabilities
You get the voice features you would expect from any modern platform. What is listed here is the layer underneath them: the part that produces something you can hand to a client, a buyer or a supervisory authority.
Build on it
Article 50 Attestation, Outbound Eligibility and Governance Gate run inside the voice application and carry its status: In controlled testing. Evidence Vault retrieval sits in the product; the production export: Working today · retested on 26 August 2026 after the deploy (measured on browser calls rather than over a telephone line). Channel Continuity covers voice hand-offs; chat and email are planned. The public API host is not available: access runs through onboarding.
Real-time validation of pricing, approved messaging and commercial commitments: the same engine, pointed at your brand rules instead of the regulation.
Read a receipt field by field and see which parts a third party can recompute without a key, and which part needs Audact's key.
Roadmap
These build on the same evidence chain.
Next step
The application on sale today is Governed Voice AI. Our engineering team configures your first governed agent for you, on the channel your clients need. You get the proof; we do the work.
The mandate layer: ceilings, scope, approvers and the refusal before an action.
The receipt explained field by field, and what a third party can recompute from it.
The one thing on this page you can buy.
Public verifier not available yet. What exists is the receipt: one per interaction, chained and hashed.
Consent provenance, scope, expiry and revocation in one store.
Every capability, with its status, in one searchable list.
