d5s
Get started
Security, EU hosting and approvals

You decide
how agents work.
At every boundary.

Choose what agents can access, which actions need approval, and where your workspace runs: in our EU cloud, in your own cloud or on-premise.

Get startedBook a demo
Kai wants to create a Linear issue

“Trial activation dropped on Wednesday” · Team Growth · Assignee Sam

DenyApprove

Permissions

Set a rule
for every tool.

Connect a tool, then choose what the agent may do with it.

A rule for every tool

Each action can run automatically, ask you first or never run. Reads run on their own by default. Writes ask first.

Approved hosts only

Agents reach only hosts approved for the workspace, project, agent or chat. Admins decide who can approve a new host.

Credentials stay outside

Connector credentials are injected outside the agent sandbox. The agent never sees the credential value.

Human approval

You stay in
the loop.

An email ready to send. A meeting ready to book. An issue ready to create. You decide what happens next.

Kai wants to create a Linear issue

“Trial activation dropped on Wednesday” · Team Growth · Assignee Sam

DenyApprove
Send the recap to 3 attendees

“Launch stays on Thursday…” · Gmail

DenyApprove
Book a 15-minute check-in, Wed 10:00

Maya, Jamie · Google Calendar

DenyApprove

Approve where you work

Cards appear in d5s, Slack or Microsoft Teams. An email links back to d5s.

Allow once

One matching action, within 15 minutes.

Allow for this run

Matching actions in this session, for at most 24 hours.

Expiry means no

A chat card works once. When it expires, the action counts as denied.

Illustrative product views. Names and data are fictional. In Ask for approval mode, sensitive actions wait for a decision. Saved policy can allow actions without individual approval; explicit denials and account permissions still apply.

Audit trail

Every change
leaves a record.

Owners and admins see security events across the organization. Filter by category, workspace and person, then export.

Data protection

Clear rules for
data and models.

Choose how model providers may use your work, for one person or the whole organization.

Zero data retention

Team and Enterprise use zero data retention by default for d5s-provided primary model routes.

No silent fallback

If no eligible no-training route is available, protected work stops. It does not move to another route.

Encrypted storage

Connector credentials, database storage and workspace files are encrypted at rest.

Model inference in the EU

ModelsWhere inference runsRetention
Claude modelsEU, where availableZero data retention
GPT modelsEU data zone, where availableZero data retention
Other modelsThe provider's regionPer route

Paid organizations in the EU use an EU route where one exists. Zero data retention applies to Team and Enterprise. The Models page in organization settings shows each model's route.

Separate controls

Four boundaries.
Enforced separately.

Access is not one switch. Deployment, model access, execution and identity have their own controls.

Secure deployment

Run in our EU cloud, in a single-tenant deployment, in your own VPC or on your own Kubernetes.

LLM provider safety

Provider API keys stay outside the agent sandbox. Model access follows your provider and routing policies. Enterprise teams can bring their own model keys.

Sandboxed execution

Agents run in isolated workspaces with controlled outbound access. Connector credentials are injected outside the agent sandbox. Workspace files can persist between runs.

Access control

Manage access through organization, workspace and project roles. Security-relevant changes are recorded for review.

Read the security documentation

EU hosting

A region is
a real boundary.

Workspace data lives in its regional cell, with its own cloud account, database, object storage, encryption keys and backups. Self-hosted and bring-your-own-cloud deployments run the same product.

Deployment regions: European Union: Serving traffic, United States: Coming soonEuropean UnionServing trafficEUUnited StatesComing soonUS

Model processing, authentication, payments and other supporting services may process data outside your selected region.

Review the subprocessors
DeploymentWhere it runsWho operates itModel keysNext step
Multi-tenant cloud, EUOur cloud, in the EUd5sd5s model routesGet started
Single-tenant cloudA dedicated d5s deploymentd5sd5s routes or your own keysTalk to us
Your own cloudYour VPC, in the region you chooseYou, with our teamYour own keysTalk to us
On-premiseYour Kubernetes, with Helm chartsYou, with our teamYour own keysTalk to us

Dedicated deployment scope and timing are agreed with your team.

Security questions

The details,
made clear.

What does zero data retention cover?

Team and Enterprise require zero data retention for d5s-provided primary model routes, including their recovery routes. Connected AI subscriptions are outside this requirement. Auxiliary services and tools have their own processing terms. Zero data retention at the model provider does not delete conversations or files stored in your d5s workspace.

Can I prevent model providers from using my data for training?

Yes. You can select no-training routes for your work, and organisation owners and administrators can require them for the organisation. Protected work stops if no eligible route is available. Training restrictions and provider retention are separate controls.

Where is my workspace data stored?

Workspace data is stored in its selected regional cell. The hosted service runs in the EU. Model processing, authentication, payments and other supporting services may process data outside that region. Regional storage does not promise EU-only model inference.

Does model inference stay in the EU?

For paid organizations in the EU, Claude and GPT models use an EU route where one exists, with zero data retention. Other models run in their provider's region. The Models page in organization settings shows each model's route.

Can I run d5s in my own cloud or on-premise?

Yes. d5s runs as a single-tenant deployment, in your own VPC or on your own Kubernetes with Helm charts. These deployments run the same product. Talk to us to agree scope and timing.

Can I use my own model keys?

Yes. Single-tenant, your own cloud and on-premise deployments can use your own model provider keys. Provider API keys stay outside the agent sandbox.

How does d5s support GDPR?

d5s processes your data as your processor under a published Data Processing Addendum. It includes EU Standard Contractual Clauses, 30 days' notice before a new subprocessor, clear deletion periods and a yearly audit right. The subprocessor list shows who processes data and where.

Which actions require approval?

Actions configured to require approval wait for a decision. Saved policy can allow actions without individual approval. Explicit denials and account permissions still apply.

Can I approve from Slack or Microsoft Teams?

Yes. Approval cards appear in d5s, Slack or Microsoft Teams. An email links back to d5s. A chat card works once, and an expired card counts as a denial.

How long does an approval last?

Allow once covers one matching action within 15 minutes. Allow for this run covers matching actions in the session for at most 24 hours. Deny stops the action.

Can I export the audit log?

Yes. Owners and admins can filter security events by category, workspace and person, then export them as CSV.

How do I close my account?

Start in Settings, under Profile. Resolve any account dependencies, verify your email and confirm closure. A 48-hour cooling-off period lets you cancel before deletion and anonymisation begin. Some records remain for documented billing, security or legal purposes. The Privacy Notice explains retention periods and exceptions.

Security resources

Report a security issue to security@d5s.tech. We keep a documented incident-response process. This page was last reviewed on 4 October 2026.