SunderSystems

Home / Data processing addendum

Data processing addendum

Questions: [email protected]. This page has not yet been reviewed by a lawyer.

Last updated: 17 November 2026

Contents
  1. Scope and roles
  2. Processing instructions
  3. Confidentiality of personnel
  4. Security
  5. Sub-processors
  6. Assistance with data subject requests
  7. Personal data breaches
  8. Deletion and return
  9. Audits and information
  10. International transfers
  11. Liability and precedence
  12. Annex I: Details of processing
  13. Annex II: Security measures

Scope and roles

This addendum forms part of the agreement between SunderSystems (“SunderSystems”, the “Processor”) and the business client that uses the platform (the “Client”, the “Controller”). SunderSystems is operated by [Owner to fill: legal entity name], [Owner to fill: registered address]. The addendum applies to personal data that the Processor processes on the Client’s behalf, and is intended to meet Article 28 of the GDPR and UK GDPR and the requirements of the Data Privacy Act of 2012 (RA 10173) for outsourcing and subcontracting of personal data processing, and the processor duties in section 541.104 of the Texas Data Privacy and Security Act.

Processing instructions

The Processor processes personal data only on the Client’s documented instructions, which are the agreement, this addendum and the Client’s use and configuration of the platform. It tells the Client if it believes an instruction infringes data protection law. The Processor does not use the Client’s data to train a model for another client. It trains a model on the Client’s data only if a user with the owner role in the Client’s workspace turns training on in settings, which the Client treats as its instruction and which is recorded as a consent. The model is trained on the Processor’s own hardware, tested against a gate before it goes live, and used only for the Client. Chat histories the Client imports are stored read-only and are never used for training, and text written by a closed model is never used for training. Details of the processing are in Annex I.

Confidentiality of personnel

The Processor ensures that people authorised to process the personal data are bound by confidentiality and have access only as needed for their role.

Security

The Processor implements the technical and organisational measures in Annex II and keeps them appropriate to the risk. The Processor holds no security certifications today.

Sub-processors

The Client gives general authorisation for the sub-processors listed on the sub-processors page, which include OpenRouter and the model provider it routes to for AI requests sent through the open-weight model route, and Xendit once payments are live. The Processor binds each to data protection obligations no less protective than this addendum and remains responsible for them. It gives notice of intended additions or replacements as described on that page, and the Client may object on reasonable data protection grounds; if the parties cannot resolve the objection, the Client may terminate the affected service.

Assistance with data subject requests

Taking into account the nature of the processing, the Processor helps the Client respond to requests to exercise data subject rights, including through the in-app export and deletion functions. If the Processor receives a request directly, it passes it to the Client and does not respond except as instructed.

Personal data breaches

The Processor aims to notify the Client without undue delay, and within 72 hours, after becoming aware of a personal data breach affecting the Client’s data. Where the Processor is the controller and the law requires it, it tells affected people and the National Privacy Commission within 72 hours. It provides the information reasonably needed for the Client to meet its own notification duties, including those to the National Privacy Commission and EU or UK supervisory authorities.

Deletion and return

At the end of the service, or on request, the Processor lets the Client export its data and then deletes it from the live system within 30 days, including any model, training data and labelling data (“Delete model and data”). Copies in backups, which are encrypted and kept on the Processor’s own disks and USB with no cloud copy, age out within about ten weeks. This applies unless the law requires the Processor to retain some of it, such as invoices, in which case it keeps that data protected and only for that purpose.

Audits and information

The Processor makes available the information necessary to demonstrate compliance with this addendum, including audit and approval logs relating to the Client’s workspace, and allows and contributes to audits, including inspections, by the Client or its appointed auditor on reasonable notice, no more than once a year unless there has been a breach, during business hours and subject to confidentiality.

International transfers

Processing takes place on SunderSystems-operated hardware in the Philippines, and with the sub-processors listed. Where processing involves a transfer of personal data from the EEA or UK that needs a transfer mechanism, the parties rely on the Standard Contractual Clauses (and the UK addendum) which are incorporated by reference, with the Client as exporter and the Processor as importer. Where the Processor transfers data onward to a sub-processor outside those regions, it ensures an appropriate safeguard is in place.

Liability and precedence

Liability under this addendum is subject to the limitations in the agreement: under the terms of service, the fees the Client paid in the 12 months before the claim arose. If this addendum conflicts with the agreement on the protection of personal data, this addendum prevails.

Annex I: Details of processing

  • Subject matter and purpose: providing the AI agent workspace, the hosted desktop and hosted models to the Client.
  • Duration: the term of the agreement, plus up to 30 days for deletion from the live system and about ten weeks for backups to age out.
  • Location: our own hardware in the Philippines, plus the sub-processors listed.
  • Nature of processing: storing, retrieving, analysing and generating text and actions using AI models, logging and deletion.
  • Data subjects: as determined by the Client, typically the Client’s staff, customers, prospects and contacts.
  • Types of personal data: as determined by the Client, typically contact details, messages, documents and other data the Client puts into its workspace, and the Client’s users’ account data.
  • Sensitive data: the Client should not put special-category or sensitive personal information into the platform unless it has agreed this with us in writing.

Annex II: Security measures

  • Encryption in transit using TLS, terminated at Cloudflare.
  • No ports open to the internet on our hosting machine: services are reached through an outbound tunnel.
  • Internal services bound to loopback only.
  • Administrative access to the hardware and the platform limited to the personnel who need it, currently the founder, with a staff audit log.
  • Audit logging of AI actions and approvals.
  • An approval step: by default a person approves before the AI sends or posts; the Client’s autonomy setting can let routine tasks run without waiting.
  • Account protections: two-step sign-in with an authenticator app, a devices and sessions page, and team invitations with roles. Email confirmation and password reset by email are built and sent through Mailgun; the mail log keeps a keyed hash of the recipient, no content, for 90 days.
  • Salted hashes for invite codes and for IP addresses in consent records.
  • Passwords stored only as hashes.
  • A separate hosted desktop container for each Client, and logical separation of each Client’s workspace data within the application, on shared hardware. Containers are Docker containers on one machine, not separate virtual machines.
  • Nightly encrypted database backups, kept for about ten weeks, on our own local disks and USB. There is no cloud copy. Desktop home folders (which can include browser sign-ins) are backed up nightly in the same way, and a deleted desktop’s copies age out in about ten weeks.
  • For security, a log of the host names (not page addresses or content) a Client’s AI desktop connects to, kept for up to 30 days.
  • AI step logs, one record per action the AI took (the tool, its inputs with known secret formats such as keys, tokens and passwords hidden and long text cut to its opening words, the result and who approved it), kept for 90 days and then deleted automatically. Free text may still contain sensitive words, so the Client should not paste secrets into prompts. Workspace owners and admins can export them before then.
  • Training on the Client’s data only if the Client turns it on, and only for the Client’s own model, which is served only to the Client. Closed-model text and imported chat histories are never used for training.
  • Open-weight models only. The route to OpenRouter tells it to use only providers that do not collect or train on prompts.
  • Developer API calls: only usage numbers (model, tokens, cost, time taken) are stored, not the prompt or the reply.