Evaluation pack
Everything on one page, so you can take it into your own meeting.
No form, no NDA, no account. Print it, save it as a PDF, forward it. Every figure on this page is read from the same registry the rest of the site reads from.
- 1What Audact is, in four lines
- 2What exists today, with the ground under every status
- 3One decision, worked out in full
- 4What a receipt establishes, and what it does not
Share
Send this to the people who have to agree.
This page is public. Nothing sits behind a form, so it can be forwarded, printed or attached to a file without anything falling away.
- Page address
- https://audact.ai/en/evaluate
1 — What Audact does
Four lines, in the order they belong in.
The problem
AI can bind your business.
In a conversation the commitment is made before anyone can review it.
The core line
Give AI a mandate. Not a blank cheque.
A mandate sets the ceiling, the scope and the expiry before the AI can act. Every action routed through the governed workflow is checked against it in the moment rather than reviewed afterwards: what crosses the line stops and goes to a person, and what does happen leaves a receipt.
The product
Control what AI can cause. Prove what follows.
Decide what an AI may cause, on whose behalf and within which limits.
What was decided and what followed from it stays recomputable, months later.
The mechanism
Authority before action. Proof through finality.
Enforced when it acts, not reviewed afterwards.
When it is questioned later, the record answers, not memory.
2 — Current state
What exists today, part by part.
Every capability below carries a status word, with the ground under it.
Voice
- voice
- Is it builtIn controlled testing
- voice
- Can I get it
Why this status
- Claim
- In controlled testing · Paid founding cohort open
- Scope
- The governed voice application for agencies: outbound and inbound calls with disclosure, policy enforcement and a receipt per governed interaction.
- Ground
- Calls run, but no paying deployment has gone through it.
- What it does not establish
- That a deployment is live with a paying customer, and that the delivery is production-hardened.
- What you can get today
- The founding cohort is open to a limited number of external agencies. Terms and figures are in the price book.
- National policy layer
- Germany: national policy layer verified. Netherlands: in progress.
- Last checked
- 26 August 2026
- Where to check it
- /voice
Compliance receipt
- compliancebon
- Is it built
- compliancebon
- Can I get it
Why this status
- Claim
- Working today · Public
- Scope
- The receipt itself and the three values a reader can recompute from it without a key. Not the export of a set of receipts, and not the publication of the Merkle roots.
- Ground
- One receipt per governed interaction; the sample files are open to anyone without an account.
- What it does not establish
- Who issued the receipt, and whether the underlying event is true.
- What you can get today
- The sample files are public. In the product the receipt sits inside a paid agreement.
- Last checked
- 26 August 2026
- Where to check it
- /proof
Receipt explorer
- receiptExplorer
- Is it built
- receiptExplorer
- Can I get it
Why this status
- Claim
- Working today · Public
- Scope
- A worked example on synthetic data. Not customer data, and not a checking service for someone else's receipt.
- Ground
- The recomputation runs in the reader's own browser on a sample file, without an account and without a key.
- What it does not establish
- It is not the public verifier. It checks the file the reader already holds, against values that file itself carries.
- What you can get today
- Free to open, nothing to buy, no account.
- Last checked
- 26 August 2026
- Where to check it
- /receipt-explorer
Production export
- bonExport
- Is it built
- bonExport
- Can I get it
Why this status
- Claim
- Working today · Paid founding cohort open
- Scope
- The production export of a set of receipts, measured on browser calls. Not a call over the telephone network, which has never been measured, and not the monthly evidence package, which is a separate artefact.
- Ground
- Retested on 26 August after the deploy: five calls, thirty validated turns, thirty receipts, across two agents and two configurations. A call from 19 August, re-exported the next day, carried its receipts as well.
- What it does not establish
- That anyone outside the platform can check the export: the key URL inside the envelope does not resolve. And the monthly evidence package, which has not been measured.
- What you can get today
- Included in a paid agreement.
- Last checked
- 29 August 2026
Control
- control
- Is it builtIn development
- control
- Can I get itInternal only
Why this status
- Claim
- In development · Internal only
- Scope
- The mandate and authority layer: ceilings, scope, approvers and the refusal in front of a governed action. The architecture is described; the runtime is being built.
- Ground
- There are no users.
- What it does not establish
- That any of it runs today, and that the figures in the interface come from a running system.
- What you can get today
- Not orderable. Design-partner conversations only, and only after the delivery gate is confirmed in writing.
- Last checked
- 26 August 2026
- Where to check it
- /control
Integrations with third-party systems
- koppelingen
- Is it builtIn development
- koppelingen
- Can I get itNot available
Why this status
- Claim
- In development · Not available
- Scope
- The integrations with third-party systems: HubSpot, Pipedrive, Cal.com, Zapier and a customer's own Slack workspace. Reading availability or a deal stage, writing an outcome back, and posting an event from the voice product into a system that is not ours. Not the six endpoints of the API itself, which have their own entry, and not the notice Audact sends itself when a form on this site is submitted.
- Ground
- Measured in the source on 28 August 2026: the code base makes exactly one outgoing call to a third party, and that is a Slack webhook with which Audact notifies itself that a form on this site was submitted. There is no client code for HubSpot, Pipedrive, Cal.com or Zapier, no key that any line reads, and no route that could receive a webhook from them: of the nine routes under src/app/api, seven take in a form and the other two are a health check and a first-party error counter. The same search does return that one outgoing call, so the absence belongs to the integrations and not to the search.
- What it does not establish
- Which integration is built first, and when. No order among them has been decided and no date has been established.
- What you can get today
- Nothing to order. A connection is scoped inside an applicable paid agreement; integrations are not priced separately.
- Last checked
- 28 August 2026
- Where to check it
- /integrations
The public API host
- apiHost
- Is it builtIn development
- apiHost
- Can I get itBy application
Why this status
- Claim
- In development · By application
- Scope
- The public API host. The code examples document the intended interface and the endpoints used within the current product.
- Ground
- The public host is not reachable yet. The endpoints used inside the product are not the same thing as a public API.
- What it does not establish
- That any individual endpoint works today. That is confirmed per endpoint, and the confirmation is open.
- What you can get today
- Production access sits inside an applicable paid agreement. Endpoints are not priced separately.
- Last checked
- 26 August 2026
- Where to check it
- /apis
The public protocol surface
- mcpHost
- Is it builtIn development
- mcpHost
- Can I get itBy application
Why this status
- Claim
- In development · By application
- Scope
- The public protocol surface: the host for the Model Context Protocol and the tools a client reaches across it. Not the REST host, which has its own entry, and not the endpoints used inside the current product.
- Ground
- Measured on 27 August 2026 with two independent resolvers: the Model Context Protocol host mcp.audact.ai does not exist in DNS. The same lookup returns records for audact.ai and staging.audact.ai and nothing for an invented subdomain, so the answer belongs to the name and not to the resolver.
- What it does not establish
- That any individual tool behind this host works today. That is confirmed per tool, and the confirmation is open.
- What you can get today
- Nothing to order. Access runs through onboarding inside an applicable paid agreement; the protocol surface is not priced separately.
- Last checked
- 27 August 2026
- Where to check it
- /mcp-server
Machine-readable marking under Article 50(2)
- art50Markering
- Is it builtPlanned
- art50Markering
- Can I get itNot available
Why this status
- Claim
- Planned · Not available
- Scope
- Marking synthetic audio in a machine-readable format under Article 50(2). Not the Article 50(1) disclosure, which is spoken in the call flow and sits inside the voice application.
- Ground
- It is not in the product today, and no date has been established. The version number that used to stand here was a release promise, not a state anyone had determined; the last nine lines carrying it were removed on 28 August 2026.
- What it does not establish
- Which release will carry it, and when. Nor that the duty is ours: under Article 50(2) it sits on the provider of the generating system.
- What you can get today
- Nothing to order and nothing to apply for. It is not part of any agreement we sell today.
- Last checked
- 28 August 2026
- Where to check it
- /compliance/eu-ai-act
Public issuer verification
- verifier
- Is it builtNot available
- verifier
- Can I get itNot available
Why this status
- Claim
- Not available · Not available
- Scope
- A service where a third party checks a receipt using the exported file plus public material alone: no database of ours, no private key, no server of ours saying it is valid.
- Ground
- No service checks a receipt on behalf of a third party today.
- What it does not establish
- Nothing. The service does not exist.
- What you can get today
- Nothing to buy and nothing to apply for. It does not exist.
- Last checked
- 26 August 2026
3 — One decision
The same worked example that stands on the home page, on Control and in the explorer.
One request against one mandate, with every figure it was measured against.
Mandate
M-204 · v3
- Principal
- Acme GmbH
- Agent
- Renewal Agent 7 · AGT-7F31
- Counterparty
- Nova Industries
- Max per commitment
- €200,000
- Discount authority
- 10%
- EMEA renewals
- 1 Jun 2026 – 31 Dec 2026
Authority position
€500,000
Mandate ceiling
- Reserved
- €300,000
- Available
- €200,000
Reserved 60%
Proposed
€250,000
Commit the Nova Industries renewal
Decision required
Authority: Escalate
€50,000 over available
Moment: 14:02:11.402 UTC
- Ground
- Request
- Value
- €250,000
- Ground
- Available authority
- Value
- €200,000
- Ground
- Excess
- Value
- €50,000
- Ground
- Mandate
- Value
- M-204 · v3 current
- Ground
- Result
- Value
- Human escalation required
Worked example. Acme GmbH and Nova Industries are fictional; no customer data appears here.
What this status establishes
- AuthorityDid the AI stay inside the configured authority and rules?EscalatedEstablished by Audact.
- QualityWas the answer correct, understandable and usable?Not assessed by AudactEstablished by the customer, not by Audact.
- OutcomeDid the promised appointment, payment or call-back actually take place?No closing source connectedEstablished by a source competent to state it: the payment rail, the scheduling system, the counterparty.
Why this status
- Claim
- Escalated · Not assessed by Audact · No closing source connected
- Scope
- One governed turn or action. Not the call as a whole, not the deployment, and not the customer relationship.
- Ground
- In an internal test a turn passed all seven configured checks, scored 100% and was recorded as allowed, while the answer it gave was not usable. The authority verdict and the quality of the answer are separate measurements.
- What it does not establish
- That the answer was correct, that the task succeeded, or that anything is compliant beyond the rules configured for this workflow.
- Last checked
- 27 August 2026
4 — Check it yourself
Three values, recomputable by anyone.
Start on the left: three values anyone holding a receipt can recompute today, without a key and without us. A fourth, inclusion in the Merkle root, follows once we publish the roots. Beside it stands what the tenant key is for and what a receipt is a record of. So the first column can be taken literally instead of on trust.
Anyone can check
The input hash, the output hash, the link to the receipt before it. Inclusion in the Merkle root is the fourth check, and it follows once we publish the roots; that publication and the open endpoint for it are in development.
No key, no account and no access to our database: the arithmetic runs on the receipt in your hand.
A receipt is a faithful record of what the system received, decided and sent. That is what it establishes, not whether the world matched it.
- What it establishes
- that the record is intact: a later change or reordering is detectable by recomputation against a retained commitment
What a receipt is about, and what it is not
Not asserted: execution, fulfilment or economic finality
Whoever can check the code can also produce it, and we hold the same key. It evidences that the record is intact, not who wrote it.
If a caller gave a wrong name, the receipt records the wrong name faithfully. Proof of process is not proof of fact, and no cryptography closes that gap.
No court has ruled on it. Say tamper-evident records, or audit-ready receipts.
- What it does not establish
- who issued it, and whether the underlying event is true
The fields this receipt carries
Input block and output block, as they stand in the sample file.
- Field
- the input hash
- Value
- 0xfeb0ca5b0546ad8c…5092dcae
- Field
- the output hash
- Value
- 0x97d47018e45e19e1…83c557d9
- Field
- the link to the preceding receipt
- Value
- 0xbfafe10889982fd2…09fe0579
- Field
- inclusion in the Merkle root
- Value
- In developmentPublication of the roots, and the open endpoint that serves them, is in development; the fourth check follows once they are published.
5 — Available now
Nothing to assemble yourself.
- Product
- Governed Voice for agencies
- Access
- Founding cohort open — first three external agencies
- National policy layer
- Germany policy pack verified · Netherlands in progress
- One-time implementation
- €1,497
- Then every month
- €499
The founding rate applies to Audact's first three external agency customers. Starting with Audact's fourth external agency customer, the standard rate is €2,995 implementation and €749 per month.
What is in it
- 5,000 minutes included, then €0.11 a minute, all in.
- The Agency Pro platform: sub-accounts per end-client, multi-tenant. Draft and test accounts are unlimited; up to five production deployments.
- The monthly evidence package for your end-client's data protection officer stays out of the offer until the production export behind it is proven — we do not sell an evidence artefact we have not tested ourselves. What stands today is the receipt, visible in the product for governed interactions.
- Primary application hosting in Frankfurt · e-mail answered within one working day.
Fourteen days to walk away.
The implementation package is fully refundable within fourteen days of signature, as long as no live call has been placed through the configuration. After that: monthly cancellation, no lock-in.
A test call does not close that window. The refund policy sets the trigger at the first production call through the deployed configuration, so the governed test call on day one and the pre-launch QA calls after it leave the fourteen days intact.
€5,090 for twelve months up front — 15% off €499 × 12. Entirely optional: monthly billing carries no annual minimum.
6 — Getting started
What you bring, what we build, and what we do not build.
What you bring
- A twenty-minute scoping call. We map your end-clients, the calls you want automated, the disclosure wording and the evidence your end-clients need each month.
- First governed test call the same day. Production go-live follows telephony, data, policy and your client's approval — and ninety days of intensive engineering support.
What we build
- Article 50 disclosure enforced at runtime — the policy engine configured for the jurisdictions you operate in.
- The evidence chain wired up: a receipt per governed interaction, visible in the product. The production export works today; the monthly evidence package is under validation — see section 2.
- Your first agent flow built end to end by our engineering team.
- Knowledge base set up — 50 documents, 200 pages — plus one CRM webhook (HubSpot, Pipedrive, Cal.com or your own).
- Pre-launch QA: 50+ test calls, checked against the disclosure and the receipt chain before you go live.
Ten things we deliberately do not build.
Every feature we skip is one less thing to defend in front of a regulator, and one less way for the evidence chain to have a hole in it. If you need any of these, pick a platform that builds them. That is a real requirement and we are not going to argue you out of it.
- Bring your own key for the model
- On-premises deployment
- A voice-cloning marketplace
- Enterprise contact-centre suites
- STIR/SHAKEN caller authentication
- Custom models and fine-tuning on your weights
- A self-serve credit tier
- Every language, counted for the count
- Squads of agents handing off to each other
- A user-tunable denoising mode
Section 2 lists what exists today, part by part.
7 — At a glance
Each line with the kind of status that fits it.
Every line below is either a document on this page, a link, or a statement we will repeat in writing on request. The status word follows the kind of thing it describes, because one word cannot carry six kinds of fact: a company is verified in a public registry, a technical control is implemented, a contract is in draft, under counsel review or contractual, a template is available or it is not, a product is in controlled testing, and a certification is held or it is not.
- Item
- IP-holding entity
- Position
- Audact AI Ltd, Companies House 17059133, London
- Status
- Verified in registry
- Item
- Contracting entity (operating company)
- Position
- AUDACT AI LIMITED, Ireland, CRO 822514
- Status
- Verified in registry
- Item
- Hosting region
- Position
- AWS Frankfurt (eu-central-1) — primary application hosting
- Status
- Implemented
- Item
- Data roles
- Position
- Audact is processor; the agency customer is controller (GDPR Art. 28)
- Status
- Contractual
- Item
- Sub-processors
- Position
- Full list on the Trust Center · 30 days' notice before a change
- Status
- Published
- Item
- Retention
- Position
- Call audio not retained by default; up to 90 days if recording is on · evidence chain per the agreement
- Status
- Implemented
- Item
- Encryption
- Position
- TLS 1.3 in transit · AES-256 at rest · crypto-shredding on erasure
- Status
- Implemented
- Item
- DPA / AVV
- Position
- Article 28-aligned; final wording pending counsel sign-off
- Status
- Counsel review
- Item
- Sample compliance receipt
- Position
- /sample-receipt.json and /sample-receipt.pdf · open, no account
- Status
- Available
- Item
- Security contact
- Position
- security@audact.ai · coordinated disclosure policy published
- Status
- Published
- Item
- Launch scope
- Position
- Voice in controlled testing · Article 50 policy: Germany verified, Netherlands in progress
- Status
- In controlled testing
What we hold, and what we do not.
Audact holds no security or AI management certification.
- Standard
- ISO 27001
- Status
- Not certified
- Ground
- No statement of applicability, no certification body engaged, no audit booked.
- Standard
- ISO 42001
- Status
- Not certified
- Ground
- No track started. The harmonised European standards are not final, so an assessment today measures a moving target.
- Standard
- SOC 2 Type II
- Status
- Not certified
- Ground
- Planned, no auditor engaged. Sequenced after the first paying deployments.
- Standard
- NEN 7510
- Status
- Not certified
- Ground
- No certification and no standards-body membership.
There is no certification under the AI Act. Nobody can hold one, us included.
Audact is not a qualified trust service provider. Its timestamps follow RFC 3161 and fix an ordering against our own chain.
Every party that touches the data, with the country it sits in.
We notify customers 30 days before adding a new sub-processor. Several of the parties are US-incorporated; those transfers run under standard contractual clauses.
Penetration testing
Not yet carried out. A first external test is scheduled before the first paying deployment; the date appears here once it is booked.
What an agency, and Audact, may say in public.
Three values can be recomputed without a key today; a fourth, inclusion in the Merkle root, follows once we publish the roots. The authentication code cannot be recomputed at all without the key, because it is keyed and we hold that key. Name the three, and say where the fourth stands.
Share
Send this to the people who have to agree.
This page is public. Nothing sits behind a form, so it can be forwarded, printed or attached to a file without anything falling away.
- Page address
- https://audact.ai/en/evaluate
