Lemo-Design

What happens when something goes wrong.

Two-step escalation

Default: notify the affected customer via the client portal and email.

Serious: also notify the Dutch DPA (Autoriteit Persoonsgegevens) within 72 hours when there is risk to people's rights — for example sensitive data type, volume, identifiability, or likely harm such as identity theft or financial loss.

Below threshold: internal written log only (date, what happened, what was done), since the AP can ask for it anytime.

Breach notifications can be filed in BacksterOS with attached tasks to track resolution.

Sensitivity note

Financial information including invoicing counts as sensitive (name + address + purchase info is identifiable).

Timeline

The 72-hour deadline for notifying the Autoriteit Persoonsgegevens starts the moment we become aware of a breach, not when the investigation is finished. If not everything is known yet, we file an initial report and add details later.

Affected customers are notified without undue delay, and at the latest within 48 hours of becoming aware, so they can meet their own obligations.

Roles and responsibilities

Remon de Vries is responsible for incident response: assessing the breach, deciding on notifications and signing off on the resolution.

Others are brought in when needed: technical support for investigation and recovery, the affected customer's contact person, the vendors involved, and legal advice for serious cases.

Detection and containment

When an incident is detected we:

• Lock or suspend affected accounts and revoke active sessions.
• Rotate exposed keys, tokens and passwords.
• Isolate affected servers or services.
• Restore clean data from encrypted backups where needed.
• Preserve logs and evidence for investigation.

Informing affected individuals

When a breach is likely to result in a high risk to people's rights and freedoms, the affected individuals are informed directly and without undue delay. The notice describes in clear language:

• What happened and which data is involved.
• The likely consequences.
• What we have done and are doing to address it.
• What they can do to protect themselves.
• Who to contact for more information.

Client data and vendors

When we process personal data on behalf of a client, we alert that client without undue delay so they can meet their own reporting duties as data controller.

Our vendors (including Hetzner, AgentMail and Stripe) are required to alert us of breaches affecting our data. We then assess the impact and follow the same escalation steps.

Post-incident review

After every incident we review what happened, why, and how it was handled. Follow-up fixes are filed as tasks in BacksterOS and tracked until completed, and the internal log is updated with the outcome.

Reporting a security issue

Found a security issue? Email info@lemo-design.com, or write to Lemo-Design, Turfmarkt 7, 8911KS Leeuwarden, Friesland, The Netherlands, or call +31 (0)58 266 42 42. We acknowledge reports and keep you informed while we investigate.

How we protect systems and data.

Access control

Access is limited to Remon only, via API key authentication. Our AI agents also have access to these systems on Remon's behalf. They can read support and WhatsApp conversations, and they may reply for us (see AI governance). Key rotation once a year is agreed (no formal written policy yet).

Infrastructure

Our servers run at Hetzner in Germany. Since 21 October 2026, servers accept SSH connections only over Tailscale, a private encrypted network. This is enforced both by each server's firewall and by the Hetzner Cloud firewall, so administrative access is not exposed to the public internet.

Backups

All servers are backed up nightly. Backups are encrypted before they leave the server and stored in a private Cloudflare R2 bucket, keeping 14 daily and 8 weekly copies. Failed backups trigger an alert. Production retention for client data also follows client project agreements (see Retention under Resources).

Encryption

In transit: public traffic is served over HTTPS (TLS). Administrative access (SSH) is only possible over Tailscale, which is encrypted with WireGuard.

Backups: encrypted with GPG on the server before upload, then stored in Cloudflare R2, which also encrypts all objects at rest with AES-256. Restores have been tested.

Server disks and databases: not encrypted at present. Disk encryption is planned.

Secrets handling

Keys, passwords and other secrets are stored in 1Password, never in code or shared documents. Access to them is limited to the people and services that need them.

Two-factor authentication

Two-factor authentication is enabled on all key accounts that give access to systems and data, including hosting (Hetzner, Cloudflare, TransIP, Hostnet), code (GitHub), payments and accounting (Stripe, Moneybird), email and support (AgentMail, Chatwoot), Meta Business and our password manager (1Password).

Updates and security patches

Servers install security updates automatically (unattended-upgrades is enabled on all servers), and other software is kept up to date.

Monitoring and logging

A Prometheus instance monitors the client portal on production. We do not use an external uptime monitor. System logs are kept on each server with size-limited rotation (about 15 to 35 MB per server); no fixed retention period is defined. Backup failures send alerts automatically.

Data minimization

We only collect and keep the data we need to deliver our services, and we delete it when it is no longer needed (see Retention and Deletion under Resources).

Working as a small team

Lemo-Design is run by Remon de Vries, who is the only person with administrative access. This keeps the number of people who can reach data to a minimum. Access for anyone else is granted only when needed, limited to what the task requires, and removed afterwards.

Vendor security

We choose established vendors and review their security and privacy documentation, including where they store data and which sub-processors they use, before using them. The current list is under Subprocessors.

How we use AI, and how we keep it in check.

Why we rely on AI

Lemo-Design is essentially a one-person company. AI is how we scale to deliver good-quality services, so a large share of our decision-making and automation runs on AI.

Not just prompting a chatbot

We don't simply type prompts into a general AI model. We run an extensive system of our own in which agents work in layers on top of other agents. That layering makes our decision-making workflows more reliable and robust than a single model answering on its own.

Tracking and oversight

A large part of our infrastructure exists to track what our AI does and to make sure it is used and governed correctly. Remon de Vries stays responsible for the work our AI agents do.

AI suppliers and infrastructure

Our agents run on Cursor, provided by Anysphere, Inc., a SpaceX subsidiary based in the United States. Through Cursor we use a wide range of AI models, from OpenAI as well as Anthropic, Google, SpaceXAI, Meta, DeepSeek, Moonshot, Z.ai and other suppliers. Data our agents process can therefore be handled by these providers, which are based in the United States and China.

Agent logs and how we keep them

Our agents keep logs of their work: the tools they call and their reasoning along the way. We keep these logs indefinitely, because we use them to train and improve our system. Before a log is stored for that purpose, we scrub out code written for clients and all personal data, such as client names, messages and other identifiers. What remains is non-personal pattern data about how our agents work. We keep it forever and say openly that we train on it.

What we train on, and what we don't

We train our AI on the filtered data we keep from agent logs: patterns in how our agents reasoned and which tools they used. It contains no code that belongs to you and that you paid for, and no personal data. We're upfront that we train on it. The code we deliver is yours. The general patterns and know-how we learn are ours.

What our AI agents can access

Our AI agents work in our own systems on Remon's behalf. They can read client conversations, such as support and WhatsApp messages, and may reply for us (see the Meta subprocessor entry and section 9 of the privacy policy).

Opting out of training

Training on scrubbed agent logs is on by default. If you don't want logs from your projects used for training, email info@lemo-design.com and we'll exclude them.

How code is written and reviewed

Our software is developed mainly by AI agents, with human review. Agents do the bulk of the development work, and a person stays responsible for what goes out. No code is pushed without a human verifying it first.

That doesn't mean a person reads every line. AI agents produce far more code than any human could review in full, so a line-by-line check of everything isn't realistic. Our approach is to review broadly: we look over the code written on your behalf, and the internal code a person writes, but we don't guarantee a complete review every time an agent changes something, down to the last character.

Instead of promising flawless code, we focus on fixing problems fast. We'd rather respond quickly when something goes wrong than claim the code will never fail. To do that, we run an extensive support system with checks and balances, built so we can quickly adjust a codebase when a problem shows up. People work in that system, and our AI agents use it too.

What you own

You get all the code you pay for. Our internal tools and paid premium services are not part of your codebase. Your project may use or benefit from them, but they remain ours and don't become your property.

Which AI does what

• Programming and code development: Cursor, with models from Anthropic and OpenAI. We only use frontier labs for code development.
• Agent support, email replies, client support and documentation: Grok, SpaceX's own AI model. Grok is not used for development.

Testing and review before deployment

Before code goes to production on your behalf, a custom process reviews and double-checks it. That process differs per project and client: basic automated checks (CI) for some, a more extensive setup when a client needs a more robust approach.

When something goes wrong

If data is leaked, we stop the process involved and notify you within 72 hours. What happens next depends on the kind of leak, so the follow-up is tailored to each case.

AI providers

• Cursor (Anysphere, Inc., a SpaceX subsidiary): AI infrastructure for code development (US). Trust center
• OpenAI: AI models for code development (US). Privacy policy
• Anthropic: AI models for code development (US). Privacy policy
• Grok (SpaceX / SpaceXAI): Agent support, email replies, client support and documentation; also available as a model through Cursor, but not used for development (US). Privacy policy
• Google: AI models available through Cursor (US). Privacy policy
• DeepSeek: AI models available through Cursor: DeepSeek V4.1 Flash (China; Cursor offers US data residency for these models). Privacy policy
• Moonshot AI: AI models available through Cursor: Kimi K2.7 Code, Kimi K3 (China; Cursor offers US data residency for these models). Privacy policy
• Z.ai: AI models available through Cursor: GLM 5.2, GLM 5.3, GLM 5.3 Flash (China; Cursor offers US data residency for these models). Privacy policy
• Meta: AI models available through Cursor: Muse Spark 1.3, Muse Wonder (US). Privacy policy

We only use the China-based providers (DeepSeek, Moonshot AI and Z.ai) when a client specifically asks for them.

Cursor may also route requests to other model providers beyond those listed here. See Cursor's subprocessor list for the full, current overview.

A data processing agreement is available on request.