Data Processing Addendum
The terms governing how Fruxon Inc. processes personal data on behalf of customers.
Last updated: September 2026
This Data Processing Addendum ("DPA") forms part of the Fruxon Terms of Service. For an individually executed copy of this DPA, please contact privacy@fruxon.com.
1. Scope & Applicability
This DPA applies when Fruxon Inc. ("Processor") processes personal data on behalf of the customer ("Controller") in the course of providing the Services under the Terms of Service.
This DPA is incorporated by reference into the Terms of Service and applies to the extent that the General Data Protection Regulation (EU) 2016/679 ("GDPR"), the UK GDPR, the Swiss Federal Act on Data Protection, the Israeli Privacy Protection Law 5741-1981 and the regulations made under it, or other applicable data protection laws require a data processing agreement.
2. Definitions
Terms used in this DPA have the meanings given in the GDPR unless otherwise defined:
"Controller" means the customer who determines the purposes and means of processing personal data.
"Processor" means Fruxon Inc., which processes personal data on behalf of the Controller.
"Sub-processor" means a third party engaged by the Processor to process personal data on the Controller's behalf.
"Data Subject" means an identified or identifiable natural person whose personal data is processed.
"Personal Data" means any information relating to a Data Subject.
"Processing" means any operation performed on personal data.
"Supervisory Authority" means the competent data protection authority.
3. Data Processing
The Processor shall process personal data only on documented instructions from the Controller, unless required to do so by applicable law. The Processor shall immediately inform the Controller if, in its opinion, an instruction infringes applicable data protection law.
Details of processing:
Subject matter: Provision of the Fruxon AI agent infrastructure platform.
Duration: The term of the agreement between the Controller and the Processor.
Nature and purpose: Processing of personal data to provide, maintain, and support the Services, including agent execution, conversation processing, and knowledge base operations.
Categories of data: Account and identity data, end-user identifiers, conversation content, agent memory, uploaded files and media, knowledge-base documents and their embeddings, execution traces, operational logs, data exchanged with systems the Controller connects, encrypted integration credentials, and any personal data contained in content processed by the Controller's agents.
Categories of data subjects: The Controller's end users, employees, and any individuals whose personal data is submitted to the Services.
4. Use of Data & AI Model Training
The Processor does not use personal data processed on the Controller's behalf to train, fine-tune, or improve any machine-learning model, whether the Processor's own or a third party's. The Processor does not sell or share such data, and does not use it for advertising or profiling.
Content the Controller's agents send to an AI model provider is governed by the Processor's agreement with that provider, or — where the Controller supplies its own provider credentials — by the Controller's own agreement with it. Where the Controller supplies its own credentials or endpoint, that provider is not a sub-processor of the Processor for that workspace.
5. Data Residency & AI Processing
Each workspace is created in a residency cell and its data at rest — the database, file storage, knowledge-base index and backups — remains in that cell's region. The cell is fixed when the workspace is created and cannot be changed afterwards.
AI processing is not bound by the cell's region
When an agent runs, the content it is working with is transmitted to the AI model providers enabled for that workspace and processed wherever those providers operate, which may be outside the cell's region. Certain models are served only from multi-region or global endpoints that route to available capacity worldwide. Residency of data at rest and residency of AI processing are separate questions, and the Processor states them separately rather than implying the first covers the second.
The Controller can constrain this. Each workspace carries an inference residency setting, enforced by the Services and configurable by the Controller in workspace settings:
Unrestricted — any AI model provider the Controller has enabled. The default.
Own cloud only — only providers running inside the Controller's own cloud tenancy (Vertex AI, Bedrock), so the residency obligation rides on the Controller's existing contract with its cloud provider. A request may still be served from a globally-routed endpoint.
In-region only — as above, and the model must be served from a regional rather than a global endpoint. The only setting under which content does not leave the region; it excludes models served solely from global endpoints.
A workspace may always be stricter than the cell it sits in, and never looser.
6. Sub-processors
The Controller authorizes the Processor to engage the sub-processors listed at fruxon.com/subprocessors. The Processor shall:
Impose data protection obligations on each sub-processor no less protective than those in this DPA.
Notify the Controller at least 30 days in advance before engaging a new sub-processor or replacing an existing one.
Provide the Controller an opportunity to object to the new sub-processor within 30 days of notification. If the Controller objects on reasonable grounds and the parties cannot resolve the objection, either party may terminate the affected Services.
AI model providers are selected by the Controller. A model provider processes personal data only for a workspace in which the Controller has enabled it. Enabling one is the Controller's authorization of that sub-processor; disabling it withdraws that authorization. Because a model provider processes nothing until the Controller enables it, the Processor making an additional provider available for selection is not a change of sub-processor and the 30-day notice above does not apply to it. The notice period governs sub-processors that begin processing without a further act by the Controller.
7. Security Measures
The Processor implements appropriate technical and organizational measures to ensure a level of security appropriate to the risk, including:
Encryption of personal data in transit (TLS 1.2+) and at rest (AES-256).
Per-workspace envelope encryption: a unique data-encryption key per workspace, wrapped by a key held in Cloud KMS that the key material never leaves. Customer-managed keys (BYOK) available for Enterprise on GCP KMS, under which disabling the customer key cryptographically erases stored credentials.
Workspace isolation enforced in the data-access layer rather than per query: every record is workspace-scoped, and a read without a workspace context fails rather than returning data.
Role-based access controls and per-agent collaborator roles (least-privilege).
Append-only credential-access audit logging, fail-closed for the sensitive tiers: where an audit record cannot be written, the credential operation is abandoned rather than proceeding unlogged. The audit table cannot be modified or deleted by the application.
Secrets are never exposed to the AI model: integration credentials appear in prompts and configuration only as placeholders, are resolved at the moment the external call is made, and are scrubbed from tool responses before the model sees them.
Deterministic controls enforced outside the model, including per-tool human approval gates, per-agent secret grant lists, read-only and sandbox integration modes, and per-agent and per-user cost caps.
Continuous dependency scanning, static analysis in the build, and CASA Tier 2 verification.
The Processor holds a SOC 2 Type 1 report and CASA Tier 2 verification. It does not hold a SOC 2 Type 2 attestation, does not currently commission independent penetration testing, and does not carry out a periodic formal risk survey, and does not represent otherwise.
For additional detail, see our Security page.
8. Retention
Business data is retained for the life of the workspace and deleted on request or when the workspace is deleted. There is no automatic age-based deletion of conversation content, agent memory or uploaded files.
Conversations, agent memory, files, knowledge base and configuration — life of the workspace.
Execution traces — by plan: 30 days (Free), 90 (Starter), 180 (Pro), 365 (Business). Enterprise is set per contract; where no value is set, a platform maximum of 730 days applies.
Credential-access audit records — not subject to the trace-retention sweep; retained for the life of the workspace and deleted with it.
Operational request logs — approximately 30 days.
Backups — the 7 most recent daily backups plus a 7-day point-in-time-recovery window, held in the workspace cell's region. Deleted data ages out within that window and cannot be removed from an existing backup individually.
9. Data Subject Rights
The Processor shall assist the Controller in responding to requests from Data Subjects exercising their rights under applicable data protection law (access, rectification, erasure, restriction, portability, objection). The Processor shall not respond to the substance of a request it receives directly, and shall promptly forward it to the Controller.
The Controller can satisfy most requests itself through the Services, which allow it to view, correct, export and delete an individual's conversations, agent memory, identity record and uploaded files. Where it cannot, the Processor will provide an export of, or delete, that individual's data within 10 business days of the Controller's written request. Where a request is complex or covers an unusually large volume of records, the Processor may extend that period once by up to a further 10 business days, by written notice given before the first period expires.
The Processor confirms in writing what was deleted or exported, and when. Deleted data remains in backups until it ages out (§8), and may persist with an AI model provider to the extent of that provider's own retention terms.
10. Data Breach Notification
The Processor shall notify the Controller without undue delay, and in any event within 48 hours after confirming a personal data breach affecting the Controller's personal data. The notification shall include:
The nature of the breach, including categories and approximate numbers of Data Subjects and records affected.
The likely consequences of the breach.
The measures taken or proposed to address the breach and mitigate its effects.
The name and contact details of the Processor's point of contact.
Information may be provided in stages as the investigation develops. The Processor will assist the Controller in meeting its own notification obligations to supervisory authorities and Data Subjects, and will preserve relevant evidence — including execution traces and conversation records — beyond the applicable retention period where it is relevant to an investigation. Notification is sent to the security contact the Controller designates; the Controller is responsible for keeping it current.
11. International Transfers
Where personal data is transferred outside the European Economic Area (EEA) or the United Kingdom, the Processor shall ensure that appropriate safeguards are in place, including the EU Standard Contractual Clauses (SCCs) or other approved transfer mechanisms. For transfers subject to the UK GDPR, the UK International Data Transfer Addendum to the SCCs applies; for transfers subject to Swiss law, the SCCs apply with the adaptations required by the Swiss Federal Act on Data Protection.
Transmission of content to AI model providers for inference (§5) is an onward transfer and is subject to this section.
Model providers process in their own jurisdictions
Each AI model provider processes content where it operates, and those locations differ. DeepSeek models are offered through Amazon Bedrock and Google Vertex AI; the Processor does not route the Controller's data to DeepSeek's own service, which stores and trains on data in China under its own terms. Where the Controller enables a provider, or supplies its own provider endpoint, whose processing takes place in a jurisdiction requiring particular safeguards, the Controller is responsible for the legal basis and safeguards for that transfer. The inference residency settings in §5 are the enforced control for constraining this. See the Sub-processor List for provider locations.
12. Israeli Privacy Law
Where the Controller is the owner of a database (בעל מאגר) under the Israeli Privacy Protection Law 5741-1981, the Processor acts as a holder (מחזיק) of that database, and this DPA is the written engagement required between an owner and a holder under the Privacy Protection (Data Security) Regulations 5777-2017.
The Controller states the security level applicable to its database under those Regulations. The Processor offers its engagement for databases classified at the basic level. As stated in §7, the Processor does not commission independent penetration testing and does not carry out a periodic formal risk survey, and therefore does not represent that it meets the additional obligations applying at the medium and high levels; an engagement covering a database at either level requires a written amendment recording the additional measures agreed.
The Controller remains responsible for the obligations those Regulations place on a database owner, including assessing the risks of the engagement before it begins, determining the security level, supervising the holder, and reporting a severe incident to the Privacy Protection Authority. The Processor reports incidents to the Controller under §10 so that the Controller can meet that obligation. Transfers of personal data out of Israel are made in accordance with the Privacy Protection (Transfer of Data to Databases Abroad) Regulations 5761-2001.
13. Audit & Information
The Processor shall make available to the Controller the information necessary to demonstrate compliance with this DPA, including its SOC 2 Type 1 report, CASA Tier 2 verification, security questionnaire responses, sub-processor list, and retention and deletion statement — available under NDA via the Trust Center. The Processor shall answer the Controller's reasonable written security and privacy questions. That information, together with those answers, is how the Processor discharges its obligation to allow for and contribute to audits, and the Controller shall rely on it in the first instance.
Where the Controller demonstrates in writing, with reasons, that this is insufficient to verify compliance in a specific respect, it may conduct a remote inspection limited to that respect — by written questions, evidence requests and, where necessary, a recorded walkthrough — on 30 days' written notice, once in any twelve-month period. An on-site inspection may be conducted where a remote inspection has been completed and has not resolved the matter, and where either a personal data breach affecting the Controller's data has occurred or a supervisory authority with jurisdiction over the Controller requires it. An inspection covers only what is relevant to the Controller's data and does not extend to other customers' data or to shared production data stores.
14. Data Deletion & Return
The Controller may delete its data at any time through the Services. Deleting a workspace runs as a 30-day grace period followed by a permanent purge. The workspace remains fully accessible throughout the grace period, so the Controller can export whatever it needs, and a reminder is sent before the period ends. Either party may cancel during the grace period; once it elapses, the purge cannot be undone.
Expiry or termination of the agreement starts that same process. At the end of the grace period the Processor permanently purges the personal data and deletes existing copies, except to the extent applicable law requires retention. The Controller may ask for the purge to be brought forward, or ask the Processor to return the data in a structured, commonly used, machine-readable format instead, within the period in §9.
For Enterprise customers with customer-managed encryption keys (BYOK), disabling the key renders every credential the Processor stores for that workspace permanently unreadable, including in backups, with immediate effect.
15. Term & Termination
This DPA takes effect on the date the Controller first accesses the Services and remains in effect for the duration of the Processor's processing of personal data on the Controller's behalf. The obligations of this DPA survive termination to the extent the Processor retains any personal data.
16. Contact
Questions about this DPA? Contact us at privacy@fruxon.com.