Does the agent export use the active version or saved draft, and when are backups deleted?

dareni · 2026-10-24T17:06:37.287Z

Can someone confirm if this exact DPA applies to a Pro account and clarify the backup deletion schedule. Also, I am building an agent and the build prompt retains “Eliott” after reloading, but a new agent export still contains the previous “Elliot” text. Does Export return the active version or the saved draft? How can I export and verify the unpublished draft, and confirm which version Test agent uses?

3 comments

· Sorted by Oldest

Niamh Byrne
·
·
Hi dareni I can see you've opened a similar ticket in our support queue. Let's continue the conversation there instead, as I might be slower to respond here.
dareni
·
·
Hi Niamh, Thank you for confirming the Self-Serve DPA applies to our Pro account and explaining the export and draft behavior. Could you please follow up on these remaining questions?
1.
Backup deletion: Has your security team provided a maximum deletion period for backup copies after manual deletion or account termination? If no fixed maximum is available, please confirm that explicitly and explain any applicable exceptions.
2.
Processing routes: For our configured openai-gpt-6-sol model, please confirm the applicable OpenAI API retention, logging, training, and regional-processing terms. Please also identify the providers and applicable data-handling terms for Google Search and Extract website content. The extraction tool’s Model input is currently set to “Let agent decide.”
3.
Client separation: On Pro, what enforced controls separate client data, tools, credentials, and retrieval access? Please distinguish project privacy from isolation between clients within the same project, including administrator access.
We are continuing public/synthetic validation only and have not approved confidential client data or sensitive-data use. Thank you, Dareni
Niamh Byrne
·
·
Hey Dareni, Backup deletion: The Self-Serve DPA doesn't set a maximum period for backup copies. When services end, backup copies are deleted under our backup and retention schedules (section 8.1), and we may keep data for as long as Data Protection Law requires (section 8.2). The DPA doesn't cover backups after manual deletion during the subscription. GPT 6 Sol: This is OpenAI's model, and without your own API key, requests are sent to OpenAI in the US. Our Security Overview states that for OpenAI models used without your own key, data isn't logged or used for training, and is covered by a DPA. OpenAI's terms: platform.openai.com/docs/guides/your-data Google Search: The provider is Serper. Extract website content: "Model" chooses the scraping service, not an AI model. The options are Apify (the default) and Browserless. "Let agent decide" means your agent picks one. If none is set, Apify is used. Client separation: Pro includes one project and two users. Roles apply to the whole project, and Custom API Keys are available to every builder in it. Asset-level permissions and RBAC are Enterprise features, so Pro can't separate clients within a project. Our subprocessor list is at trust.relevanceai.com/subprocessors.

Niamh Byrne
·
·
Hi dareni I can see you've opened a similar ticket in our support queue. Let's continue the conversation there instead, as I might be slower to respond here.
dareni
·
·
Hi Niamh, Thank you for confirming the Self-Serve DPA applies to our Pro account and explaining the export and draft behavior. Could you please follow up on these remaining questions?
1.
Backup deletion: Has your security team provided a maximum deletion period for backup copies after manual deletion or account termination? If no fixed maximum is available, please confirm that explicitly and explain any applicable exceptions.
2.
Processing routes: For our configured openai-gpt-6-sol model, please confirm the applicable OpenAI API retention, logging, training, and regional-processing terms. Please also identify the providers and applicable data-handling terms for Google Search and Extract website content. The extraction tool’s Model input is currently set to “Let agent decide.”
3.
Client separation: On Pro, what enforced controls separate client data, tools, credentials, and retrieval access? Please distinguish project privacy from isolation between clients within the same project, including administrator access.
We are continuing public/synthetic validation only and have not approved confidential client data or sensitive-data use. Thank you, Dareni
Niamh Byrne
·
·
Hey Dareni, Backup deletion: The Self-Serve DPA doesn't set a maximum period for backup copies. When services end, backup copies are deleted under our backup and retention schedules (section 8.1), and we may keep data for as long as Data Protection Law requires (section 8.2). The DPA doesn't cover backups after manual deletion during the subscription. GPT 6 Sol: This is OpenAI's model, and without your own API key, requests are sent to OpenAI in the US. Our Security Overview states that for OpenAI models used without your own key, data isn't logged or used for training, and is covered by a DPA. OpenAI's terms: platform.openai.com/docs/guides/your-data Google Search: The provider is Serper. Extract website content: "Model" chooses the scraping service, not an AI model. The options are Apify (the default) and Browserless. "Let agent decide" means your agent picks one. If none is set, Apify is used. Client separation: Pro includes one project and two users. Roles apply to the whole project, and Custom API Keys are available to every builder in it. Asset-level permissions and RBAC are Enterprise features, so Pro can't separate clients within a project. Our subprocessor list is at trust.relevanceai.com/subprocessors.