The model providers we use
We treat AI providers the same way we treat any other sub-processor: we select them on capability, security posture and data-handling guarantees, and we list them publicly on the sub-processors page.
myflow currently routes AI workloads to one or more of the following, depending on the feature and your account configuration:
- OpenAI — via the OpenAI API (enterprise data-handling terms).
- Anthropic — via the Anthropic API for Claude models.
- Google — via the Gemini API.
The set of providers may change as the AI landscape evolves. Material changes to sub-processors are announced via the sub-processor page; you can subscribe to changes there.
We do not train models on your data
myflow does not train, fine-tune, or otherwise build models on customer content. We use AI providers under their API terms, which — by default across all three providers above — do not train their general models on data sent through their APIs. We do not opt in to any data-sharing arrangement that would change this.
If we ever build features that fine-tune a model on a specific customer's own data for that customer's exclusive benefit, we will only do so with that customer's explicit, opt-in consent and clear scope.
What gets sent to AI providers
AI features only send the data they need. Concretely:
- The prompt assembled by the feature you're using (your instructions, the relevant context the feature has pulled together).
- The model parameters (which model, temperature, etc.).
- No bulk export of your account. No background scraping of unrelated data.
Different features use different amounts of context. A "rewrite this paragraph" action sends one paragraph; a "draft an email to this lead" action sends the relevant lead fields. We aim to send the minimum needed to produce a good result.
Logging: monitoring and quality, not surveillance
We log inputs and outputs of AI calls for two reasons:
- Quality and debugging. When a feature produces a wrong, broken or strange result, we need to be able to reproduce it.
- Abuse and safety monitoring. Detecting misuse of AI features in a way that would harm other customers or third parties.
AI logs are subject to the same protections as the rest of your data: stored in the EU, encrypted at rest, access-controlled. They are not shared with the model providers beyond the original API call, and they are not used to train any model. Logs are retained for a limited period and then deleted.
Model output is not authoritative
Generative AI can be wrong, biased, or out of date. myflow's UI surfaces AI output as a draft — something to review and edit, not to ship blindly. We design AI features to put humans in the loop on anything that leaves your account (emails sent, content published, billing actions). When we ship features that automate work on your behalf, we make the boundaries and the off-switch obvious.
Your controls
- AI features can be disabled at the workspace level. Disabled features make no AI calls at all.
- Workspace admins control which team members can use AI features.
- If you need to confirm exactly what was sent to a provider on a given call, our security team can help — contact us.
Responsible-use commitments
We commit to:
- Naming the providers we use, publicly, on this site.
- Telling customers ahead of time when a provider changes or a feature changes how it uses AI.
- Not training models on your data without explicit opt-in.
- Designing AI features to keep humans in the loop on consequential actions.
- Treating AI logs with the same care as the rest of your data.