Data protection, in the open
GDPR, EU data residency & your data
AgentWorks is built for European teams that treat GDPR, the EU AI Act and data residency as requirements, not nice-to-haves. Here is exactly how we handle your data — and who we work with.
- Fully hosted in the EU
- Zero-retention model contracts
- DPA available on request
01 — Commitments
What we commit to
Fully hosted in the EU
AgentWorks runs on our own infrastructure in Germany (Hetzner) — application, database, authentication and file storage all live there. Large language model calls use EU endpoints where the provider offers them.
Zero-retention model contracts — the guarantee
The guarantee we build on is architectural: EU hosting plus model providers engaged on a no-training, zero-retention basis. That is what actually keeps your content from being retained or used for training — not a text-scanning layer, which can never be fully watertight.
Optional PII-detection layer
An additional layer can detect and mask common personal-data patterns before content reaches a model. It reduces exposure — we do not market it as a guarantee, because free-text detection always has a miss rate.
Audit trail + human oversight
Every state-mutating action is recorded to an append-only audit log, and high-risk actions can require human-in-the-loop approval before they run.
02 — Sub-processors
The vendors behind the platform
The third parties that process data on our behalf. We work with each on a Data Processing Agreement and inform customers of material changes.
| Sub-processor | Purpose | Region | Data-processing terms |
|---|---|---|---|
| EU infrastructure hosting (self-hosted) | Infrastructure hosting for the application (frontend & backend) and database — self-hosted on our own servers | EU (Germany) | Infrastructure provider; data processed under our own operational control |
| AWS (Bedrock) | Large language model inference — hosts Anthropic (Claude) and OpenAI open-weight (GPT-OSS) models | EU (eu-north-1) | No-training, zero-retention |
| Google Cloud (Vertex AI) | Large language model inference — hosts Google (Gemini) models | EU | No-training, zero-retention |
| OpenAI | Large language model inference (GPT chat models) & text-embedding generation | US (direct API — no EU-hosted OpenAI endpoint in production today) | No-training, zero-retention |
| Mistral | Large language model inference | EU (France-domiciled provider) | No-training, zero-retention |
| Stripe | Payment & subscription processing | EU + US | PCI-DSS certified (Stripe) |
| Transactional e-mail service | Transactional email delivery | EU | Covered by DPA |
Optional integrations
Present in the platform but only active where explicitly configured. Listed for completeness — ask us for current production status for your account.
| Sub-processor | Purpose | Region | Data-processing terms |
|---|---|---|---|
| xAI (Grok) | LLM inference — optional fallback provider | EU | Configured per account |
| Perplexity | Web-search tool called by agents (not chat inference) | EU | Configured per account |
| Error monitoring service | Error monitoring | EU | Configured per account |
A complete, named sub-processor list is available under our DPA — request it via contact.
03 — Security
Built to keep tenants apart
How the platform itself is built to keep tenants apart and credentials safe.
Tenant isolation by design
Every database query the backend makes is explicitly scoped to your organization at the application layer, not left to a database policy alone. Realtime updates (live chat/task streams) get their own row-level security scoped to what each person is actually allowed to see.
Encryption of credentials & keys
Integration tokens and platform model-provider keys are encrypted at rest. Nothing is stored as plain text.
Security headers by default
The application ships a Content-Security-Policy, HSTS, and standard anti-clickjacking / MIME-sniffing headers.
Dependency hygiene
Automated dependency scanning runs in CI, with Renovate/Dependabot keeping libraries current. Critical vulnerabilities block merges.
04 — Retention
How long we keep data
If you close your workspace subscription
A 30-day grace period follows workspace closure, after which a purge sweep anonymizes your tenant record. (This is separate from the personal "delete my account" action above, which is immediate — see Data subject rights.) Money-ledger and audit-log rows are retained for roughly 7 years to meet standard bookkeeping rules — they stay linked to the anonymized record, not to identifying information.
Operational log retention (policy)
Our data retention policy sets chat-interaction logs at 30 days, dispatch/action audit logs at 6 months, and the compliance audit trail at 7 years. Chat messages themselves are never automatically deleted. Automated enforcement of these windows is rolling out.
05 — GDPR
Our GDPR commitments
Data Processing Agreement (Art. 28)
AgentWorks acts as processor and your organisation as controller. Request a DPA using the button below and we will work through it with you directly.
Data subject rights
Every user can self-service their own access, portability and erasure rights from Settings — a one-click data export (Art. 15/20) and immediate account deletion (Art. 17: your data is anonymized and wiped on confirmation, not scheduled for later), no ticket required. Rectification is a normal profile edit. Anything else (e.g. a request about someone else’s data) we handle per-request by email.
Records of processing (Art. 30)
We maintain a record of processing activities, available to supervisory authorities and to customers under their DPA.
Breach notification
Our commitment is to notify affected customers without undue delay after becoming aware of a personal-data breach.
International transfers
Where a sub-processor processes data outside the EU, we cover the transfer with Standard Contractual Clauses via the provider’s own DPA.
Privacy policy
Our privacy policy covers cookies and how to control them in your browser — see it for full detail.
06 — EU AI Act
EU AI Act: ready, not a blanket claim
No platform is automatically "EU AI Act compliant" — the Act classifies risk by use case, so whether a given agent is high-risk depends on how you deploy it (for example, HR screening or credit scoring can be high-risk). AgentWorks gives you the controls the Act expects, so you can deploy responsibly:
- Per-agent risk classification
- Append-only audit trail of every action
- Human-in-the-loop approval on high-risk actions
- Article 50 transparency — clear AI-interaction disclosure
- An optional PII-detection layer to reduce exposure
Certifications roadmap
Today we operate on infrastructure fully hosted in the EU (Germany), zero-retention model contracts, a published sub-processor list, and a privacy policy; a DPA is available on request. We are evaluating third-party security certifications (for example SOC 2 or ISO 27001) as we grow into larger enterprise and agency deployments. Talk to us about your specific assurance requirements.
07 — FAQ
Frequently asked questions
Where is my data stored?
AgentWorks is fully hosted in the EU on our own infrastructure: the application, database, authentication and file storage all run on servers we operate in Germany, and large language model calls use EU endpoints where the provider offers them. The guarantee we rely on is EU hosting plus zero-retention model contracts — not a redaction layer. An optional PII-detection layer can further reduce what reaches a model, but it is a complement to that guarantee, not a substitute for it.
Do you train AI models on my data?
No — the model providers we use are engaged on a no-training, zero-retention basis for API traffic, and that architectural guarantee is what we rely on, not a redaction promise. Free-text personal-data detection can never be 100% complete, so we do not market our optional PII-detection layer as a guarantee that personal data never reaches a model — only that it reduces exposure.
Do you sign a Data Processing Agreement (DPA)?
Yes — under GDPR Article 28, AgentWorks acts as processor and your organisation as controller, and we work through a DPA directly with business customers on request, covering any non-EU sub-processor transfers via Standard Contractual Clauses.
Is AgentWorks EU AI Act compliant?
AgentWorks is AI Act-ready rather than blanket "compliant" — whether a specific use case is high-risk depends on how you deploy the agent. We provide the building blocks the Act expects: per-agent risk classification, an append-only audit trail, human-in-the-loop oversight for high-risk actions, and Article 50 transparency (making clear when you are interacting with AI).
Who are your sub-processors?
Our core sub-processors are Hetzner (EU infrastructure — application and database, self-hosted on our own servers in Germany), AWS Bedrock and Google Vertex AI (EU-hosted LLM inference for Anthropic, OpenAI open-weight, and Google models), OpenAI (US-hosted GPT chat models and text embeddings) and Mistral (LLM inference), a transactional email service, and Stripe (payments). See the sub-processor table above — including optional integrations — for the full category breakdown, and request our DPA for the complete named list.
How do I exercise GDPR data subject rights?
Access and portability (Art. 15/20): download your own data any time from Settings → Account. Erasure (Art. 17): delete your account from the same page — your personal data is anonymized and your content wiped immediately on confirmation, not scheduled for later. Rectification is a normal profile edit. For anything else, contact us and we handle it per-request.
Talk to us
Need a DPA or security review?
Request our Data Processing Agreement, the sub-processor list, or a security questionnaire — or start free and see the compliance controls in the product.