Trust by design
Trust begins with a chain you can follow.
When AI takes part in processes, requirements, knowledge and decisions, promises are not enough. Organisations need to know: Where is the data? Which models are working? Who has access? How is AI labelled? What is logged? And which responsibility stays with people? The answers are gathered here in one place.
Not with a seal at the end. This page shows where your data is held, which tool touches it, who acts – and what of that can later be proven.
Data
The normal case starts in Switzerland.
Application, database, file storage, backups and mail delivery are held at a Swiss provider. Where part of your data leaves this area, there is a nameable reason – and you find it in the next chapter.
The platform is provided by Traniva AG, based in Hittnau, canton of Zurich. Operations run with a Swiss provider whose data centre is in Lucerne.
According to the provider, the data centre is in central Switzerland and data stays in Switzerland throughout. Mail gateways are spread across Rothenburg, Rümlang and Zurich, name servers across the Swiss plateau. Access control, surveillance, CO₂ fire protection and round-the-clock monitoring come on top.
| Area | Location |
|---|---|
| Application | Switzerland |
| Database | Switzerland |
| File storage | Switzerland |
| Backups | Switzerland |
| Mail delivery | Switzerland |
| AI automation n8n Cloud | EU · Frankfurt |
| Language model OpenAI | provider |
| All services in detail → | |
Operating and processing locations can differ by product, contract and provider. The details that govern a specific enterprise contract are recorded contractually.
Two legal orders, one level of protection
Switzerland is not part of the EU – for data protection that is not a drawback but an additional layer.
Swiss FADP
The revised Federal Act on Data Protection applies directly, because provider and operations are in Switzerland.
EU GDPR
For customers in the EU, processing is additionally governed by the General Data Protection Regulation – as stated in our terms.
Adequacy
The European Commission recognises Switzerland as providing an adequate level of data protection. A transfer to Switzerland therefore needs no additional safeguards such as standard contractual clauses.
Swiss platform operation
The Swiss hosting infrastructure used by Traniva is subject to Swiss law. For external AI or automation services, the legal and transfer mechanisms applying to those providers and to the respective contract configuration apply in addition.
Roles in the contract
Whether Traniva acts as controller or processor depends on the use and is settled in a separate agreement where required.
What this does not replace
A location alone is not compliance. What matters remains the configuration, the permissions and the context of use at the customer.
How customers are kept apart
What the belonging rests on
Not on a setting somebody has to maintain. Acquiring a team licence creates a key of its own for that organisation – automatically, with nothing to do.
Held on the workspace
Every workspace carries the belonging as a detail of its own. If someone moves to another organisation, their earlier workspaces do not travel with them – they stay where they belong.
For applications too
An interface access can be bound to the same belonging. A key for one organisation reaches no workspace of another, whatever it asks for.
AI
AI is an additional path – not where your work is stored.
Your models are held in Switzerland. Only when you start an agent does an extract leave that path: through the automation layer in Frankfurt to the model provider. Without an agent run this does not happen.
Data flow
What happens during an agent run?
Not the whole environment is passed on, only the content and context needed for that particular run.
What you enter – and what that means outside Enterprise
What you enter is your decision
The content of a request comes from the user. What is written into a prompt or handed over as process text is decided solely by the organisation using TAOM – and it is transmitted to the model provider.
Traniva does not screen this content for personal data, trade secrets or third-party rights. Screening would only be possible if we read along – and that is precisely what we do not do.
Why this matters below Enterprise
Below the Enterprise level the standard configuration applies: provider, model and processing region are preset and cannot be agreed individually. For anything you disclose through a prompt, that provider’s terms then apply.
Recommendation: outside Enterprise, do not enter personal data, secrets or anything under special supervision into prompts. If you need that, choose Enterprise – there provider, hosting and region are laid down contractually, up to operation in your own environment.
How the request runs and what the provider states
What is sent
- What is sent the process text and its context – not the entire environment
- Purpose Analysis, generation, classification
- Storage and region follow the provider and its contract
What the provider states
According to OpenAI’s published information, data from the programming interface is not used to train the models by default. Inputs and outputs may be retained for a limited period for abuse monitoring.
For eligible API customers, OpenAI additionally offers zero data retention (ZDR) and regional data processing for supported configurations. These properties do not apply automatically to TAOM standard licences. They can be agreed as part of an Enterprise configuration, provided they are technically available for the models and endpoints used and enabled by the provider.
Which settings apply to your contract is confirmed in writing – we do not claim it here as a blanket statement.
With Enterprise you choose
Provider, model, hosting and processing are by agreement – including a model in your own environment or a service you already run. This is also set out in the licence terms.
That allows AI processing to sit in the same legal system as the rest of your data.
What the model makes of it
How the respective model provider technically processes inputs and which retention rules apply follows from that provider’s service and from the configuration agreed for TAOM. Traniva documents to the customer which provider is used, which data is transmitted and which contractually confirmed properties apply to the specific configuration.
For the preset models below the Enterprise level, that provider’s contractual and processing terms therefore apply. What follows from them – including in data protection terms – depends on those terms and on the applicable regulations, not on an undertaking by Traniva.
What Traniva can state: which provider is used, what is transmitted, and that credentials stay server-side. With Enterprise configurations, provider, model, hosting, processing region and retention mode can additionally be laid down contractually – up to running the model in your own environment.
Control
An agent is given permissions. Not a blank cheque.
Identity, permission, action, human approval – in that order. A machine can act, but it cannot be accountable.
Responsibility
What is the agent accountable for?
Capability
What can it carry out?
Knowledge access
Which information may it use?
Tools
Which systems may it reach?
Authority
What may it decide on its own?
Approval and output
What needs human approval, and which artefacts may it create or change?
For people
Identity → role → permission → action
For agents
Agent identity → role → authority → action → evidence
On its own
The agent acts within clearly set limits without asking.
With approval
The agent analyses or proposes – a person approves.
Never
The action is not permitted to the agent.
It has to stay answerable: who acted, in which role, with which rights, on which object, when and with what result.
Who is accountable for what
TAOM
The platform provided and the documented functions.
AI provider
Its own services under its own contract terms.
Customer
Processes, permissions, context of use and business approvals.
User
Their own authorised actions.
Agent
Carries no legal responsibility. Its authority is defined organisationally.
And therefore
Approval never sits with the agent – otherwise nobody would be accountable.
No AI in critical applications in regulated industries
The draft of the European Annex 22 on artificial intelligence limits its use in GMP-critical applications to static, deterministic models. TAOM uses language models and is expressly not intended for such applications.
Process modelling itself is not affected. Where TAOM is used in a regulated environment, AI-assisted elaboration remains a proposal that passes through your existing approval path – which is exactly how it is built.
Evidence
Not just working securely. Being able to explain later what happened.
A certificate confirms that certain criteria were examined. It does not replace what has to come first.
Every file and every document carries an origin that can be checked without access to TAOM. That is the difference between an assurance and a record.
Provenance seal on the model
Every file produced carries origin, version, time, licence state and AI label – verifiable against changes by a SHA-256 checksum with a secret key (HMAC). A foreign origin is never adopted.
Document seal on documents
Invoice, licence document, data processing agreement, certificate and contract documentation carry a number, a seal and a QR code – verifiable at taom.ai/beleg, without access to TAOM.
Four states
- Created by people
- AI GENERATED – produced by TAOM AI
- AI MODIFIED – existing artefact edited in TAOM
- Imported (unknown) – external file whose origin TAOM does not claim
What is kept on the artefact
- Source
- Timestamp
- TAOM version
- AI marker
- Editing history
- Licence status
- Version
Status per feature – in use, partly organisational, planned
| Area | Status |
|---|---|
| Encrypted transport | in use |
| Credentials server-side, never in the browser | in use |
| Identities and permissions | in use |
| Tenant separation | in use |
| Logging – operational log, default 30 days, configurable per environment | in use |
| Encrypted storage | partly organisational |
| Backup and restore | in use |
| Network protection | in use |
| Updates and incident handling | partly organisational |
“Partly organisational” means: the technical basis is in place, the evidence additionally depends on a procedure at the customer – who reviews, when and against what.
Where TAOM can support – and where not
Where TAOM can support
- Data protection
- AI governance
- Internal control system
- Process documentation in GxP settings
- Auditability
- Traceability
- Enterprise governance
What that does not mean
Whether a specific use fully meets a particular regulation depends on configuration, process, organisational measures and the regulatory setting.
On the EU AI Act: TAOM provides mechanisms for transparency, traceability, human oversight and documentation – provenance, roles, authority limits, logs and versioning. There is no certification under the regulation.
For regulated industries
Our evidence chain follows the expectations placed on computerised systems in regulated environments: timestamps in coordinated universal time, tamper-evident chaining via checksums, and attribution to the acting account through a non-speaking identifier – traceable in house, anonymous outside.
Deletions are recorded in full. Recording of creation and modification is currently being added; the status is stated in the audit report.
Enterprise
When standard is not enough, data processing becomes part of the contract.
Five routes to more sovereignty. Which one applies is examined per undertaking, enabled technically and confirmed in writing – none of it applies automatically.
Standard configuration: no ZDR assurance. Enterprise: ZDR possible by agreement and technical approval. Availability and scope are examined for provider, model, endpoint and region and documented in the Enterprise agreement.
What Traniva documents for each Enterprise configuration
- the model provider in use
- the models and endpoints used or permitted
- the place or region of processing
- the retention mode enabled
- the properties assured by contract
Options include a model in your own environment without any outbound call, a Swiss provider for model processing, open models such as Apertus, developed at ETH Zurich and EPFL, or your existing contract with a provider you already use. TAOM is built so the model can be exchanged – it is a component, not a prerequisite.
Documents
Everything procurement and data protection need – in one place.
For a security review you need documents, not promises. Contracts are the evidence layer of what is described above – here they sit together, with version number and status.
Evidence you have to be able to find
A record is only useful if it can be found when an audit comes. Because review obligation, risk, control and test result sit on the process step, the question can be asked across all models: where is something still missing – before somebody else asks?
Which records exist and what they say
| Evidence | What it says | Status |
|---|---|---|
| Person certification | Three levels – Modeller, Agents, Infrastructure – each with a verifiable number | in use |
| AI provenance on the artefact | Created by people, AI-generated, AI-edited or imported | in use |
| Validity seal on the model | Held on the model and verifiable by its number | in use |
| Document seal on documents | Invoice, licence document, data processing agreement and certificate – verifiable at taom.ai/beleg | in use |
| Contract documentation | The wording of the agreements at the time of purchase, with version numbers and consent timestamps | in use |
| Reviewed by people | Approval via a ticket in the delivery system; status changes report back to the model through our own Atlassian app | in use |
| Method compliant | The model follows the TAOM rules: roles as lanes, questions at gateways, TAOM shapes | partly organisational |
| External security certification | Assessment by an independent body | planned |
What security reviews ask
Is customer data used to train public models?
In the current TAOM standard configuration, API content is not used for training under the published terms of the model provider in use. For differing Enterprise configurations this follows from the provider and operating model agreed. The configuration applying to the customer is documented in writing.
Where is data stored?
Depending on the operating and contract model. The governing region is recorded in the enterprise contract.
Can TAOM be used without AI?
Yes. Process Studio can be used entirely without an agent run – modelling, importing, enriching and exporting work without a single call.
Can an agent decide on its own?
Only within explicitly defined limits. Review, validation and approval stay with people.
Is TAOM certified under the EU AI Act?
No. TAOM provides mechanisms that support transparency, traceability, human oversight and documentation. There is no external certification.
Is TAOM GxP validated?
Not as a blanket statement. Whether a specific use is validated depends on the system and process context at the customer.
Verify rather than believe
Do not trust the claim. Examine what it rests on.
TAOM aims to make it traceable how data, people, agents, models, decisions and artefacts work together. If you need further documents for a security review, procurement, data protection or compliance, we put them together for your specific setup.