AI FEATURES TERMS
Version 1.0 · Effective 18 September 2026 · Part of the Agreement (Terms of Service Section 6; DPA Section 5)
Convenience translation; the Ukrainian text prevails.
Part A. Screen text (shown before the first AI node is added)
Before you add an AI step
AI steps — nodes, agents, vector search, the editor assistant — send data to a language model. Here is what you need to know.
Where the data goes.
- Oxta key (default): requests run on Cloudflare Workers AI — the same infrastructure the platform runs on. Data does not leave Cloudflare's perimeter. Models:
glm-5.3-flash(text),embeddinggemma-300m(vectors). We do not grant Cloudflare permission to use request content to train models. - Your own key: requests go directly to the provider you specify, under your contract with it. We do not control how it stores or uses them.
What exactly is sent. Your prompt with values substituted from previous steps, the node's system instruction, and for search — the query text and fragments of your dataset.
What is not sent. Your e-mail and name, credentials and keys (redacted automatically), payment history, data from other flows.
Important: if you place your customers' personal data in a prompt, it will be sent to the model. You are responsible for having a legal basis. Health, biometric, belief and criminal-record data must not pass through the platform.
What we store. The model's request and response are written to run history like any other step and kept under the same rules (365 days and up to the 10,000 most recent runs per flow). There is no separate switch for AI steps alone, but content storage can be switched off for a whole flow in its settings.
Limitations. The model can be wrong or make things up. Output is not professional advice. Reproducibility is not guaranteed. Verify results before use.
Your control. You can withdraw this confirmation in settings → "Documents & consents". After withdrawal you cannot add new AI steps, and already-configured ones stop running: the flow fails at the AI step without sending a request.
By confirming, you as data controller instruct us to send data to language models as described above and authorise Cloudflare Workers AI as a sub-processor.
☐ I have read and understand what data is sent to models
[ Confirm and add AI step ] [ Cancel ]
Part B. Full text
1. What AI features are
1.1. AI features are: AI nodes in flows (text generation, classification, extraction); agents; vector search in datasets, including converting uploaded text into vectors; the editor assistant and help search.
1.2. Data is sent to language models only when these features are used. Flows without AI steps send nothing to models.
2. Two modes and the responsibility boundary
2.1. "Oxta key" mode (default). Model calls run via Cloudflare Workers AI. Cloudflare is our sub-processor (DPA Annex 2); data stays within the Cloudflare infrastructure the whole platform runs on. Models as of this version: @cf/zai-org/glm-5.3-flash (text, agents, assistant) and @cf/google/embeddinggemma-300m (vectors). We may swap models for functionally equivalent ones within Workers AI with notice on the sub-processor page. Operations count against plan quota.
2.2. "Own key" mode. You provide a key and endpoint for an OpenAI-API-compatible provider. Requests go directly to it and do not consume quota. Responsibility boundary: we are responsible for sending the request unaltered (except secret redaction) only to the endpoint you specified and to no one else; the provider is responsible, under your contract with it, for storing, using and protecting requests. This is a transfer on your instruction (DPA Sections 2.3, 5.3, 6.5).
3. What is and is not sent
3.1. Sent: your prompt with substituted values from prior steps; the node's or agent's system instruction; for vector search — the query text and dataset fragments; for the assistant — your question and the flow structure.
3.2. Not sent: the account owner's e-mail and name; credentials, keys and passwords stored on the platform (redacted automatically); payment history; data of other flows and other accounts.
3.3. Personal data in prompts. The platform does not filter prompt content. If you place personal data in a prompt (a customer's name, message, order) it will be sent to the model — that is your instruction as controller. You are responsible for the legal basis and for informing data subjects. Special categories are prohibited (Terms Section 5.3).
4. What we store
4.1. The model request and response are stored in run history as a step result — with the same access and retention as other steps (365 days and up to the 10,000 most recent runs per flow; DPA Annex 1). No separate switch exists for AI steps alone; you may switch content storage off for a whole flow (DPA clause 7.5), delete a single run, or delete the flow with its history.
4.2. Dataset vectors are stored in Cloudflare Vectorize until the dataset is deleted.
5. Limitations
Model outputs may be inaccurate, incomplete or fabricated; are not legal, medical, financial or other professional advice; may differ on re-run with identical input. You are responsible for verifying results and for actions your flow takes on them (sending messages, changing records, etc.). Our liability is limited per Section 14 of the Terms.
6. Confirmation and withdrawal
6.1. Confirmation is requested once before the first AI step is added and recorded in consent records with this document's version, time and location (editor). A substantive change to this document = new version = renewed confirmation.
6.2. Legal nature. The confirmation is a documented controller instruction to the processor (Article 28(3)(a) GDPR) and prior written authorisation of the Cloudflare Workers AI sub-processor (Article 28(2)). The basis for processing your own data as a user when using the assistant is contract performance (Article 6(1)(b)). The basis for processing your customers' data is determined by you as controller.
6.3. Withdrawal. You may withdraw in settings → "Documents & consents". Effects: (a) adding new AI steps is blocked immediately; (b) execution of already-configured AI steps stops immediately — a flow that reaches an AI step fails at that step and no request is sent to any model; the rest of the flow runs as usual; (c) requests sent before the withdrawal cannot be recalled; results stored in history remain until you delete them.
6.4. Withdrawal does not affect the lawfulness of processing before it.
7. Changes
Changes follow Section 17 of the Terms. A change of AI sub-processor follows DPA Section 6 (30 days, right to object).