Intramedial est en bêta restreinte. À l’inscription ou à la connexion, vous devez fournir le mot de passe de bêta-testeur que vous avez reçu ou en demander un à preview@intramedial.com. Merci.
How the EU AI Act applies to Intramedial, which part of the Service is an AI system, how Number One meets the transparency rules, and the Article 25(4) agreement for customers who build Intramedial into a high-risk AI system.
This page sets out how Regulation (EU) 2024/1689, the EU AI Act, applies to Intramedial, as amended by Regulation (EU) 2026/1744, the Digital Omnibus on AI. It covers our own role under the Act, the one AI system we operate, and what we owe you if you build Intramedial into an AI system of your own.
Section 5 is a binding agreement. You accept it when you create an account, together with the Terms of Service, the and the . The other sections describe the Service and do not add obligations.
Intralock · Baux · Parley · Fleet Map
A Trustvice Product
Les agents agissent quand le plan dit oui, et le registre dit pourquoi.
Faites passer actions d’agents, hash de contenu, contrôles d’API, requêtes DNS, webhooks, appels LLM et commits git par Intramedial. Obtenez un reçu signé vérifiable par tous – sans nous croire.
The Service is provided by Avonao, Inc., 254 Chapman Rd, Ste 208 #21918, Newark, Delaware 19702, USA ("we", "us") (legal address). Our mailing address, for general correspondence only, is Avonao, Inc. • 2261 Market Street PMB 10671, San Francisco, CA, 94114, United States. It is not our legal address: all legal correspondence must be sent to the legal address in Delaware stated above (see the Imprint).
In plain language: Intramedial coordinates, gates and records the work of AI agents, but it is built from rules, not from a model. It is not an AI system and it is not your agent. Number One, our assistant, is an AI system, and it tells you so. If you use Intramedial with a high-risk AI system, Section 5 is the written agreement the AI Act asks the two of us to have.
The Agent Coordination Plane is not an AI system. Each of its modules carries out rules we wrote: check this agent's identity, admit or refuse this action against its lease and mandate, require a second administrator for this change, append this entry to the audit chain, compute this Merkle root, sign this bundle, verify this signature against this key. The admission decision, made by the component we call Intralock, is a pure function of the lease, the mandate, the policy and the request. It makes no network call and uses no model. None of the modules infers from its input how to produce predictions, content, recommendations or decisions. Recital 12 of the Act says that systems based on rules defined solely by people to execute operations automatically are not AI systems.
Modules that govern and record AI systems are controls and records, not part of the AI system. Several modules act on, or keep records about, the agents you run:
What the module does
Modules (groups)
Know which agents exist, who owns them, and which are unsanctioned
Fleet and world model (agent identities, owners, keys, shadow-agent discovery)
Decide what an agent may do and for how long
Intents and leases; Mandates and packs
Stop, hold or require a human at the point of action
Enforcement points (including quarantine, halt and dual control)
Test agents and detect problems before and after release
Detection and rehearsal
Settle disputes between agents and their operators
Negotiation and courts
Record what happened and let you recover from it
Record and recovery (the audit chain, the witnessed ledger, signed ledger bundles)
Your agent decides what to do; these modules decide whether it may, and record that it did. We are not the provider or the deployer of your AI system because it uses them. An admission record, an audit-chain entry or a ledger bundle is not a conformity assessment, does not make your system compliant, and does not by itself show that an obligation under the Act, such as record-keeping (Article 12), transparency to deployers (Article 13), human oversight (Article 14) or transparency to people (Article 50), is met. It is evidence you can use when you show how you meet them.
Two points deserve a word:
Human oversight controls help you do Article 14 and Article 26 work; they do not do it for you. Dual control, halt and quarantine are controls you configure. Whether a person with the right competence, authority and information is in place, and whether they actually use the control, is your responsibility as provider or deployer.
The ledger records, it does not classify. A ledger bundle shows which entries existed in your workspace's audit chain at the time it was sealed. It does not show that your system is lawful, accurate or safe, and it is not the technical documentation that Annex IV requires of a provider.
Number One is an AI system. Number One answers questions about Intramedial on the website and in the dashboard, where it can also look up your workspace when that is switched on. It is built on general-purpose AI models from Anthropic, PBC (see Sub-processors), and we put it into service under our own name, so we are its provider under Article 3(3). It is not a high-risk AI system: it is not used for any purpose listed in Annex III, and it is not a safety component of a product covered by Annex I. In the dashboard it is read-only: it cannot change anything in your workspace. For workspaces in the United States it can read records of the workspace; for workspaces in other regions it answers from the documentation only.
What we are not. We do not place a general-purpose AI model on the market, and we do not train models on your data. Because we provide no high-risk AI system and no general-purpose AI model, the duty to appoint an authorised representative in the Union (Articles 22 and 54) does not apply to us. Our EU representative for data protection purposes is named in the Privacy Policy.
Article 50 has applied since 2 August 2026. Number One was put into service under it, so no transition period applies to it.
You are told it is an AI system (Article 50(1)). The chat panel labels Number One as an AI assistant at the top, where you see it before your first message, in the clear and distinguishable way Article 50(5) requires. In the dashboard the panel also says that it can be wrong and that it only reads.
Its answers are marked in a machine-readable way (Article 50(2)). Every answer Number One generates, on the website and in the dashboard, is sent with the HTTP response header AI-Disclosure: mode=ai-originated, from the AI-Disclosure header draft (draft-abaris-aicdh). Software reading the response can tell that the text was generated by an AI system without parsing it.
Where the marking does not reach. Text you copy out of the chat panel does not carry the header with it. Carrying it would mean hiding a watermark inside the words, which reformatting removes and which would alter text you may want to quote exactly. Article 50(2) asks for marking as far as technically feasible; this is where that ends.
Requests about you are not answered by the model. A question that is really a data-subject request is met with a fixed reply and a support ticket, with no model call.
Messages written by people and shown by a program, such as an error or a limit notice, are not AI-generated and carry no marking.
4. If you use Intramedial with your own AI system#
Using Intramedial does not make your AI system compliant. Classifying your system, building its logging, managing its risks and assessing its conformity are your obligations, and we cannot take them on. What Intramedial gives you is a record of what your agents were admitted to do and did, in an audit chain that can be sealed into a signed bundle that anyone can check without trusting you or us. The Compliance pack page explains how customers export it, including for Article 12.
Dates. Under the Digital Omnibus, the obligations for high-risk AI systems apply from 2 December 2027 for systems listed in Annex III and from 2 August 2028 for systems covered by Annex I.
Keep logs for long enough. Article 19 (for providers) and Article 26(6) (for deployers) require automatically generated logs of a high-risk AI system to be kept for at least six months, unless other law says otherwise. We delete records when they leave your plan's retention window, which is described in Retention and erasure. Team (365 days) and Scale (730 days) keep records for at least six months, and so does Enterprise unless your contract sets a shorter window. On every other plan, including Pro+ at 180 days, export your records and keep them yourself. A ledger bundle you have kept stays verifiable offline with our published keys after we delete our copy.
Article 25(4) of the AI Act applies when a third party supplies tools, services, components or processes that are used or integrated in a high-risk AI system. That third party and the provider of the high-risk system must agree in writing on the information, capabilities, technical access and other assistance the provider needs to comply with the Act, based on the generally acknowledged state of the art.
If you use or integrate any part of Intramedial in an AI system that is high-risk under Article 6, you are that provider and we are that third party. This section is the written agreement between Avonao, Inc. and you, the customer who accepted it.
We make it part of every account, not only of accounts that tell us their system is high-risk. We cannot see how your system is classified, and an obligation we owe should not depend on our guessing correctly.
We supply the Intramedial modules (agent identity and fleet records, leases and mandates, admission and enforcement points, detection and rehearsal, negotiation and courts, and the audit chain, witnessed ledger and signed ledger bundles), the records they produce, the published public keys, the offline verifier, the REST API, the MCP server, the integrations and adapters we ship, and the dashboard. We do not supply an AI model, other than the assistant described in Section 2.
On written request, at no extra charge, we will provide:
a description of the intended purpose, capabilities and known limitations of each module you have integrated;
the documentation of the interfaces and of the record and bundle formats, including the meaning of each field and which fields record what we observed and which record what you or your agents told us. Much of this is already published in the Record Practice Statement, Verifying ledger bundles and Record Format Versions;
what each module records about your workspace, the fields recorded and the retention applied to them;
the security measures that protect the modules and the signing keys, as described in Key management and on the security page;
notice of any change to a module you use that could reasonably affect your compliance, before the change takes effect. A change made to fix an active security incident is the exception; we will tell you about it as soon as it is safe to.
You can retrieve the records we hold for your workspace through the API and the dashboard, and export them in a machine-readable format. You can verify a ledger bundle yourself, offline, without contacting us, using our published keys with our verifier or your own code. That stays true after your account ends: Continuity describes how.
We will give you reasonable assistance and, at your direction, assist a notified body, a market surveillance authority or an auditor dealing with your system, including by answering documented technical questions about the modules you use.
We will not, and cannot, carry out your conformity assessment, sign your EU declaration of conformity or decide how your system is classified. Those acts are the provider's alone.
This agreement does not make us the provider or the deployer of your AI system, does not move any of your obligations to us, and does not change the role either of us has under the Act. If you put our name or trade mark on a high-risk AI system, or substantially modify one, Article 25(1) may apply on its own terms; neither of us can contract out of it.
This agreement runs for as long as you use any Intramedial module in your AI system. After your account ends, we will keep answering requests under 5.3 about the modules you used, for as long as the Act requires you to keep your documentation. We keep your records only as long as the Terms of Service say, so export the ones you need before then.
This agreement supplements the Terms of Service and the Data Processing Addendum. Where it conflicts with them on a matter Article 25(4) governs, this agreement prevails.
We change this page when the Act, its guidance or the Service changes, and update the date at the top. When you create an account, we record the version of this page you accepted: its date and a fingerprint of its text. For a material change to Section 5 we tell account owners by email at least 30 days before it takes effect, as we do for the Terms of Service.
A Trustvice Product
