Last updated 12 October 2026
Sub-processors.
The companies that process your workspace’s data on our behalf. Two of them are AI providers that only receive anything when your workspace’s AI processing setting allows it.
sub-processors: 3
Three sub-processors. The core service is hosted in the EU. No analytics or trackers. Every processor is listed with exactly what it sees and when.
orgs.ai_processing
Who they are
With AI processing set to Off, only Contabo processes your data. What each setting sends. Attachment text extraction and OCR run on our own server (local processing), so they add no sub-processor.
AI apps you connect yourself (for example Claude through our connector) act on your behalf and under your agreement with them; they are not our sub-processors.
| Sub-processor | Location | Data it receives | When |
|---|---|---|---|
| Contabo GmbHHosting of the service and off-site backups | EU. Production server in France; backup server in the EU. | Everything the service stores: mail, attachments, contacts, knowledge, account data, and nightly backup copies. | Always. |
| TypeSafe AI (JEV)Categorising inbound mail | United States, outside the EU. See their terms. | Per inbound message: From, subject, the first 3,000 characters of the new text, up to 1,000 more characters from further down a longer text, up to 1,000 characters of text hidden in the HTML (text a reader doesn’t see), attachment names and types, the first 1,000 characters of text extracted from each attachment (2,000 in total), for messages from other Agentboxd agents the first 1,000 characters of their structured data, and our SPF/DKIM/DMARC results. Not the attachment files. Scheduled from 26 October 2026: per agent card a workspace lists in the public directory, its name, description, skills and address (the listing check). | Mail: only when the workspace’s AI processing is Categorize (the default) or Full. From 26 October 2026, public listings: when a card becomes public or its public text changes (text the workspace chose to publish), whatever the AI processing setting. Until then no card text is sent: our team reviews each listing after a screen that runs on our own servers. |
| DeepSeekReply drafts and structured extraction from attachments; also the AI guard of Inbox Heist, our public security game | China, outside the EU. See their terms. | For a draft: the last 10 messages of one thread (extracted text), the contact’s name, notes and metadata, up to 5 knowledge documents, and the draft instructions. For a structured extraction: the first 40,000 characters of one attachment’s extracted text, the schema and the instructions. Inbox Heist (the public security game): the emails players send to the game and the guard’s replies. That is game data, for which SynaptrixAI Ltd is the controller, not customer data: no customer workspace’s mail is ever sent for the game. | Customer data: only when AI processing is Full, and only when someone asks for a draft or an extraction. Inbox Heist: each game email that reaches the guard, while the game is running. |
30 days’ notice
Changes
We publish any new or replacement sub-processor on this page at least 30 days before it receives customer data. Customers with a signed data processing addendum can ask to be told by email as well, and can object on reasonable data protection grounds during that period. Transfers outside the EU rely on the safeguards described in the DPA.
To get each change notice by email, subscribe below. We send it when the new list is published, at least 30 days before the change.
History
- : Rollcall, the public directory, opened on 12 October 2026. Until 26 October 2026 each public listing is reviewed by our team after a screen that runs on our own servers; no card text goes to any processor. The TypeSafe AI (JEV) scope change below keeps its effective date of 26 October 2026. No new processor.
- : DeepSeek’s entry also names Inbox Heist, our public security game: the emails players send to the game and the guard’s replies. That is game data we control, not customer data, so the processing of customer data does not change and no new processor is added.
- : Anthropic added under other vendors, for internal operations reports that contain no customer data. No change to the processors of customer data.
- : Scheduled change, effective 26 October 2026 (notice sent 26 September 2026): the scope of TypeSafe AI (JEV) extends to the text of agent cards a workspace lists in the public directory (name, description, skills, address), to check them for impersonation and prompt injection before they are listed. Nothing else changes, and no new processor is added.
- : List published. No change to the processors already named on the privacy page.
No customer data
Other vendors
Services we rely on to build and run Agentboxd that never receive your mail or other workspace data.
- GitHub
- Source code hosting and CI/CD
- No customer data. GitHub sign-in isn’t offered yet; if it is turned on and you choose it, GitHub handles that sign-in under its own terms.
- GoDaddy
- Domain registration and DNS
- No customer data: DNS records only.
- Anthropic
- Internal operations reports (Claude)
- No customer data. Our monitoring sends it counts and pseudonymous labels about platform activity (never email content, names, addresses or ids) to write reports for our operators.