Klex Cloud Data Processing Agreement

1. Parties and scope

This Data Processing Agreement ("DPA") forms part of the Klex Cloud agreement ("Agreement") between stagewise GmbH, Obernstraße 50, 33602 Bielefeld, Germany ("stagewise", "we"), and the business customer identified for the relevant Klex Cloud organization ("Customer", "you"). It takes effect when incorporated into that organization's Agreement under the Klex Cloud Terms of Service. No separate signature or DPA-specific acceptance is required. Each organization's Agreement remains separate. The DPA version incorporated at acceptance applies; publication of a newer version does not itself amend an existing Agreement.

This DPA applies to personal data we process on your behalf to provide Klex Cloud, including hosted Klex Bots, Klex Machines, APIs, integrations and related support ("Customer Personal Data"). For self-hosted Bots, it covers only processing we actually perform through Cloud or agreed support, not infrastructure you operate independently. It does not cover stagewise IDE.

You act as controller or, where you process data for another controller, as its authorized processor. We act as your processor or subprocessor accordingly. You must have the relevant controller's authority to engage us and issue instructions.

GDPR terms have the meanings in Regulation (EU) 2016/679. Applicable EU and Member State data-protection law governs processing under this DPA; other mandatory law remains unaffected. A "Subprocessor" is a processor we engage to carry out part of this processing. A "Personal Data Breach" has the meaning in Article 4(12) GDPR.

Our independent processing for purposes such as our own account administration, billing and separately defined analytics is described in the Privacy Policy, not authorized by this DPA. Actual purposes determine the role: instructed content copied into support or diagnostic systems remains protected under this DPA. Controller-side retention does not authorize unrelated retention or reuse of that content.

Contact privacy@klex.bot for data-protection matters and support@klex.bot for service, return or deletion requests.

2. Processing and instructions

Annex 1 describes the processing. You determine its purposes and are responsible for a lawful basis, appropriate notices, permissions and lawful instructions. You configure users, workloads and connected-system permissions appropriate to your use. These responsibilities do not excuse our own obligations, including enforcement of agreed access and approval restrictions.

We process Customer Personal Data only on documented instructions, including the Agreement, this DPA, configurations and actions of authorized users, and additional documented instructions. Bot actions within an authorized workload do not authorize independent use by stagewise.

If Union or Member State law requires other processing, we inform you of the requirement before processing unless the law prohibits this on important public-interest grounds. We immediately inform you if, in our opinion, an instruction infringes the GDPR or other Union or Member State data-protection provisions. The parties address the concern before the affected processing continues, except where lawful preservation or other mandatory action is required.

We do not use Customer Content to train our models, independently exploit Bot content or memory, or conduct unrelated product research. Necessary hosting, execution, storage, transformation and routing to deliver the instructed service remain permitted. Customer-agreed troubleshooting does not authorize training or unrelated use by us or our Subprocessors.

Additional instructions outside the agreed service scope may require agreement on implementation. No new assistance or audit fee is created by this DPA, and commercial arrangements must not obstruct mandatory obligations.

3. Confidentiality, security and debugging

We ensure that persons authorized to process Customer Personal Data are bound by confidentiality or an appropriate statutory obligation. Access is limited to authorized responsibilities and the information needed for them. Confidentiality duties continue after access or the Agreement ends.

We implement appropriate technical and organizational measures under Article 32 GDPR, taking account of the state of the art, implementation costs, the processing and risks to individuals. Annex 2 describes the agreed measures. We assess their effectiveness regularly and may update them without materially reducing overall protection or overriding an expressly agreed obligation.

Full-content Bot debug tracing is off by default. Only technical staff enable it, following discussion and agreement with you, for a limited time and the agreed purpose. Such access is audit-logged. Traces can contain complete conversations, sensitive information and LLM inputs and outputs. They are retained for no more than 30 days and can be deleted earlier on request. Other operational telemetry follows its separately applicable purpose and retention; it must not be used to bypass these limits for full-content traces.

Temporary support debugging may involve necessary excerpts in the support and issue-management tools in Annex 3. Our own Bots may analyze agreed debug traces using the troubleshooting inference providers listed there. The same agreement, purpose limits, access logging, Subprocessor obligations and transfer protections apply. Information shared is limited to what is necessary for the investigation.

API keys and integration credentials are not intentionally included in logs or traces. This is not a guarantee that accidental inclusion can never occur or that all sensitive content is automatically redacted. Necessary security or legally required processing must have its own lawful justification and is not permission for unrelated use.

4. Model providers and connections

Runtime model access may use a provider account/API key you supply or be included in an expressly agreed Cloud offering. Unless such an offering includes model access, you supply the necessary access under your own provider arrangement. This DPA does not itself include model access or guarantee any particular provider's availability. You may also connect messaging channels, MCP services and other tools. Relevant content, credentials, instructions and tool inputs/outputs are transmitted as needed for the authorized connection. Transformation of content does not necessarily make it anonymous.

A provider's role depends on the actual arrangement, not simply who pays or supplies an API key. Providers we engage to perform our processing obligations are subject to section 5. A provider you engage independently operates under your applicable arrangement; this does not remove our duties for processing and transmission we perform. You must ensure your instructions and connections are lawful and suitable for the workload.

No promise is made that every independently selected provider has the same training, retention or location settings. This section does not reclassify stagewise-engaged troubleshooting providers as your independent providers.

5. Subprocessors

The Customer grants general written authorization for the Subprocessors and processing functions identified in our publicly available Subprocessor List, which is incorporated into this DPA by reference. The list identifies the Subprocessors authorized at the time the Agreement takes effect. Subsequent additions or replacements are governed by the notification and objection procedure in this section.

Before adding or replacing a Subprocessor, we notify your designated contract or privacy contact by email of its identity, relevant function, processing locations and transfer safeguards sufficiently in advance to give you a reasonable opportunity to object on data-protection grounds. Updating the public list does not independently authorize an addition or replacement without following this procedure.

We work with you to resolve a timely objection before the disputed processing begins, including considering an alternative arrangement. If no reasonable solution is available, the parties address discontinuation or termination of the affected service under the Agreement and mandatory law before the disputed processing begins. This process does not permit unilateral changes to unrelated contractual rights.

We bind each Subprocessor by a written contract imposing the applicable data-protection obligations required by Article 28(4) GDPR, including sufficient security guarantees. We remain fully liable to you for performance of those obligations as required by that Article. Necessary information about these arrangements is available on request, subject to lawful confidentiality protections.

6. Locations and international transfers

Primary hosting of Cloud-managed Bots and Machines is in Germany. The specified Grafana Cloud Bot debug-trace storage has Germany data residency. These statements do not promise that all service traffic, support access, security processing or onward processing is confined to Germany or the EU/EEA. Other provider processing depends on the relevant service and applicable arrangements identified in Annex 3.

We carry out a transfer restricted by Chapter V GDPR only on documented instructions and under a lawful transfer arrangement. This may include an applicable adequacy decision covering the recipient and data, or appropriate safeguards such as the European Commission's international-transfer Standard Contractual Clauses, together with required assessments and supplementary measures. EU storage or a provider's public statement alone is not proof of coverage.

We provide information about relevant processing and access countries and applicable safeguards on request, subject to lawful redactions. Material changes follow section 5 or the Agreement's amendment requirements, as applicable. If a relied-upon mechanism ceases to apply, we establish an appropriate alternative or suspend the affected transfer until it can lawfully resume.

Mandatory transfer clauses prevail over conflicting Agreement or DPA terms. This DPA does not itself purport to complete a particular set of international-transfer clauses for an unidentified transfer. Separately applicable clauses and their required details must be validly incorporated for the relevant arrangement.

7. Assistance and personal-data requests

Taking account of the nature of processing, we assist you through appropriate technical and organizational measures, insofar as possible, in responding to requests under Chapter III GDPR. If we receive a request concerning Customer Personal Data, we notify you without undue delay and do not independently decide its substance unless instructed or legally required. We may direct the requester to you and verify identity or authority proportionately without disclosing data to an unauthorized person.

Taking account of the nature of processing and information available to us, we assist with obligations under Articles 32–36 GDPR, including security, breach notifications, data-protection impact assessments and prior consultation. Your use of the Services does not transfer your own controller obligations to us or remove our processor obligations.

8. Personal Data Breaches

We notify affected Customers directly by email without undue delay after becoming aware of a Personal Data Breach affecting their Customer Personal Data. Notice is sent to an appropriate designated contract, privacy or security contact, or an authorized organization owner where no separate contact has been designated. Please keep contact details current; we take reasonable steps to reach an authorized contact where needed.

To the extent available, notice describes the nature of the breach, affected categories and approximate numbers of individuals and records, likely consequences, response and mitigation measures, and a contact for further information. Missing information may follow in stages without undue further delay. Notification is not postponed until an investigation is complete.

Our DevOps personnel coordinate response, containment and remediation, preserve relevant evidence and cooperate with you. Public status updates may supplement but do not replace direct notice. You determine notifications you must make to individuals and authorities, without restricting legal duties independently imposed on us. No fixed hourly notification or resolution guarantee is created, and notice is not itself an admission of liability.

9. Compliance information and audits

We make available information necessary to demonstrate compliance with Article 28 GDPR and allow for and contribute to audits, including inspections, by you or your mandated auditor.

Available documentation and independent assurance, where available, may be used first if they adequately address the inquiry, but do not replace further verification where necessary. Routine audits are coordinated with reasonable notice, confidentiality, security and protection of other customers' information. These arrangements must not make verification ineffective or delay urgent examination following a breach, credible compliance concern or authority request. No absolute annual limit or certification-only substitute applies.

We maintain applicable processing records and cooperate with competent supervisory authorities as required by law.

10. Return, deletion and residual copies

At your choice, we return or delete Customer Personal Data after the processing services end and delete existing copies unless Union or Member State law requires storage. We also comply with valid deletion instructions during the Agreement. Return, switching and retrieval rights under the Agreement and mandatory law remain preserved.

Contact support@klex.bot to arrange secure return or deletion without requiring continued account access. We may verify authority proportionately. Return scope, formats and any applicable switching timelines follow the relevant service arrangement and mandatory law. Any remaining retrieval requirement is coordinated with your instructions before deletion is initiated, not used to postpone an already triggered deletion maximum.

When a Bot or Machine is deleted, its Customer Personal Data is deleted within a maximum of 30 days after that resource's deletion. When an organization is deleted, its data and all its Bots' data are deleted within a maximum of 30 days after organization deletion. If organization deletion requires managed Machines to be deleted first, each Machine remains subject to its own 30-day maximum measured from its deletion.

These maxima include backups and copies held in Grafana, support/debugging systems and other systems within stagewise's processing chain, including relevant Subprocessor copies. Copies awaiting deletion remain protected and restricted to the necessary recovery, protection or deletion process. If a backup is restored, applicable deletion instructions are reapplied before affected data returns to ordinary use. Earlier action is taken where law requires it.

These commitments do not purport to erase copies independently held by providers outside our processing chain under your separate arrangements. Deleting a user account does not by itself instruct deletion of shared organization data. Stopping execution, pausing a resource or exhausting credits is not the same as deleting it. A credit-related minimum period before discretionary deletion is not a deletion SLA or permission for indefinite retention.

Customer Personal Data that Union or Member State law requires us to retain is appropriately restricted to that purpose, protected under this DPA and deleted when the requirement ends. Independent controller-side financial or legal recordkeeping does not justify retaining an entire live workload. We confirm completion of instructed deletion on request.

11. General terms

This DPA continues while we process Customer Personal Data, including residual copies. It prevails over inconsistent general Agreement terms concerning instructed personal-data processing, subject to mandatory law and applicable transfer clauses. Individually agreed terms prevail for their subject without overriding mandatory protections. Covered switching matters remain subject to the switching addendum and mandatory rights.

The Agreement's liability provisions apply only to the extent lawful and consistent with mandatory data-protection and transfer-clause obligations. Nothing limits mandatory data-subject rights or liability, removes responsibility under Article 28(4) GDPR, or changes statutory burdens of proof. No certification, uptime SLA, fixed patch deadline or recovery-time guarantee is created by this DPA.

Amendments follow the Agreement's amendment provisions and applicable law. Publication, silence or continued use alone is not deemed acceptance of changes to contractual rights. The Subprocessor process in section 5 is distinct from an amendment of this DPA. The Agreement's governing-law and jurisdiction provisions otherwise apply, without prejudice to mandatory remedies, supervisory powers and transfer-clause provisions.

Annex 1 — Processing details

Item	Description
Subject and purpose	Klex Cloud hosted Bots and Machines, memory, files, execution, APIs, authorized integrations and enabled voice connections; necessary instructed security operations and agreed support/debugging.
Nature and frequency	Receiving, recording, organizing, storing, consulting, executing, transforming, transmitting, exporting, restricting and deleting data, continuously or intermittently according to use.
Data flows	Messages through Klex Chat, supported platform connectors or Customer-configured MCP reach a self-hosted or Cloud-hosted Bot and relevant content is sent to the runtime model provider used under section 4, through Customer-provided access or an expressly agreed Cloud offering. Enabled voice connections may additionally use LiveKit. Agreed troubleshooting may send relevant trace content through stagewise's own Bots to the inference providers in Annex 3.
Duration	During instructed service provision, any lawful return/retrieval process and deletion. Section 10's resource/organization deletion maxima and section 3's full-content trace retention apply.
Individuals	People represented in authorized workloads, potentially including users, employees, contractors, customers, prospects, suppliers, business contacts and others whose data the Customer lawfully makes available. Actual categories depend on the workload.
Data types	Depending on use: names and contact details; account, organization and workload identifiers; messages, prompts, model responses, memory, files, code and configuration; linked credentials; tool inputs/outputs and metadata. Voice can include audio, room/participant information, browser-call display names and network/connection metadata. Traces may contain the same content.
Sensitive information	Workloads and traces may contain sensitive personal information. Article 9 or 10 GDPR data requires the Customer's applicable legal authorization and safeguards appropriate to the risk. This DPA does not establish a legal basis or guarantee suitability for every sensitive workload. Service restrictions remain applicable.
Customer authority and instructions	The Customer and organization identified in the Agreement; authorized users/configuration and agreed support communications supply instructions under section 2.

Annex 2 — Technical and organizational measures

• Encryption and credentials. Managed production customer-data paths and storage use encryption in transit and at rest, including databases, files, Bot/Machine disks, backups and applicable Grafana traces. Customer API keys and integration credentials are stored on each Bot's encrypted AWS EBS hosting volume. No separate application-level encryption is implied. These descriptions do not promise encryption for arbitrary Customer-configured endpoints or independently self-hosted infrastructure.
• Personnel access. Production customer-data access requires MFA, Cloudflare Zero Trust, explicit allowlisting and sufficient permissions. Access is recorded in infrastructure and/or product audit logs. Authorized personnel are subject to confidentiality and purpose restrictions.
• Isolation. Each Bot runs in an independent container; each Machine is isolated at KVM virtual-machine level. Application permissions enforce user, organization and Bot access boundaries. Containers and virtual machines are distinct isolation measures, not identical guarantees.
• Vulnerability management. GitHub Dependabot and regular internal checks support vulnerability identification. Fixes are deployed promptly to managed infrastructure, and open-source Bot updates are made available for customer-managed deployments. Customers manage updates to their independently self-hosted software. No fixed remediation deadline is promised.
• Recovery. Daily backups are taken for managed Bot and Machine data, and restoration is tested periodically. No fixed restoration-test frequency, recovery-point or recovery-time guarantee is promised. Backup deletion remains subject to section 10.
• Debug controls. Full-content tracing is off by default, staff-enabled only after Customer agreement, time-limited, purpose-restricted and audit-logged, with 30-day retention and earlier deletion available. Credentials are not intentionally logged; comprehensive automatic redaction or impossibility of accidental inclusion is not promised.
• Incidents and review. DevOps personnel coordinate incident handling and direct notification under section 8. Measures are assessed regularly for effectiveness, including through vulnerability checks and periodic restoration testing. No unagreed security certification is implied.

Annex 3 — Authorized Subprocessors

Our authoritative, current Subprocessor List is maintained publicly in the Klex Trust Center and incorporated into this DPA by reference.

The list identifies our authorized Subprocessors, their processing functions and relevant processing locations, with additional identity, contact and transfer-safeguard information made available as required by applicable law.

We maintain dated records of Subprocessor List versions to document which providers were authorized under each Customer's Agreement.

Additions and replacements are subject to the advance notification and objection procedure in section 5. Updating the public list alone does not authorize a new or replacement Subprocessor without following that procedure.

Customer-selected providers independently contracted by the Customer are not automatically stagewise Subprocessors. Providers processing data solely for stagewise's own controller purposes are not covered by this authorization.