Legal
Subprocessors
All legal pages · Terms · Privacy Policy · Acceptable Use · Cookies · DPA · Subprocessors · Report a security problem
These are the companies that process data for Digital Minions, what each one does, and what it receives. Each client workspace uses only the ones its features need. An owner can see which are in use on Settings › Security › Compliance.
We checked this list against the code on 27 October 2026: every outside service the app calls, and every key it reads. We will give 30 days' notice of a new subprocessor, by email to owners and on this page.
Core
| Company | What it does for us | What it receives |
|---|---|---|
| Railway | Hosting: runs each client's app, its database and the private browser and render workers | Everything stored for that client |
| Anthropic (Claude) | AI model for minions' work; reads uploaded PDFs; Claude web search | The words of each request: instructions, the business knowledge and customer details the task needs, tool results |
| OpenAI | AI model (when chosen), images, voice-note and meeting transcription, search embeddings (when chosen) | Request text; images asked for; audio sent for transcription; knowledge text for search. Sent with store: false on the Responses API |
| Google (Gemini) | AI model (when chosen), search embeddings (when chosen), Gemini web search | Request text and knowledge text for search |
| Twilio | Phone numbers, calls, text messages, WhatsApp messages, call recordings and carrier registration (A2P 10DLC) | Phone numbers, message text, call audio and recordings, the business's registration details |
| Resend | Email: sign-in links, alerts and the emails a person approves; receiving forwarded email | Email addresses and message content |
| Stripe | Billing: plans, invoices, metered usage and payments | The client's billing details and usage amounts. Card details go to Stripe directly, never to us |
| xAI (Grok) | Nell's voice for phone calls (xAI realtime voice), spoken voice replies, meeting and call transcription; AI model, images and video when chosen | Call audio and transcripts, Nell's instructions and approved business facts, the text of voice replies, meeting and call audio for transcription, request text. Sent with store: false on the Responses API; API requests are kept 30 days for abuse checks and never trained on |
| Anthropic (web search and fetch) | Claude's web search and web page reading, when a minion on Claude looks something up on the web | Search queries and web addresses only. Never a person's email address, phone number or street address (checked in code) |
| OpenAI (web search and the search safety check) | OpenAI's web search, when a minion on an OpenAI model looks something up on the web; and the safety check (moderation) on each search query we send, whichever search runs it | Search queries only. Never a person's email address, phone number or street address (checked in code) |
| xAI (web search and X search) | Grok's web search, and its X search when a job is about posts on X, when a minion on a Grok model looks something up | Search queries only. Never a person's email address, phone number or street address (checked in code) |
| EMBL-EBI (Europe PMC) | Reads research papers (abstracts and open-access full text) when a minion researches | A paper's public identifier (PubMed, PMC or DOI) only; nothing about the business or its people |
| NCBI (PubMed E-utilities) | Reads a paper's abstract when Europe PMC can't | A paper's public identifier (PubMed id) only; nothing about the business or its people |
| arXiv | Reads a preprint's abstract when a minion researches | A preprint's public identifier only; nothing about the business or its people |
| Google Fonts | Typefaces for the app's pages | The visitor's IP address and browser details, as with any web request. No cookies |
Only when that feature is set up
| Company | What it does for us | What it receives |
|---|---|---|
| Meta (Messenger, Instagram) | Replies to the business's Facebook Page and Instagram messages | Message text and the sender's Meta ID |
| Brave Search | Web search and Find my business | Search queries (at most 300 characters) |
| TypeSafe (Jev) | Picks the right minion for a job when work is handed out, and checks a short excerpt when a business fact is filed or a draft gets a second look: whether it is still current, and whether it is about money or a legal promise. It never writes messages | A short excerpt of the request, fact or draft being checked, with codes, passwords and card numbers taken out. Not used for health or law workspaces. TypeSafe keeps an operational record of the request |
| Deepgram | Voice-note and meeting transcription (when chosen) | Audio. Every request is sent with Deepgram's model-improvement opt-out, so Deepgram keeps no audio or transcript after answering |
| Recall.ai | The meeting bot that joins a call when the client sends a minion to a meeting | Meeting audio (and video when recorded); it records only, we transcribe the audio ourselves, and we ask Recall.ai to delete the media afterwards |
| Anthropic (Claude Managed Agents) | Hands and Minion View: a cloud computer a minion works in, when an owner turns it on | The task, the files it needs and the secrets an owner gives it (kept in Anthropic's vault) |
| Amazon S3 (or an S3-compatible store) | The nightly off-site backup copy, kept in the workspace's region | A full copy of the client's database, locked to the client's key only when the client vault is set up |
| Fivetran | Read-only reporting pipes (Connect your data): brings in ad (Meta, Google Ads) and online-store numbers and history from the apps the client connects on Fivetran's own page, into our reporting store on Amazon Web Services | What those apps hold for the reports (ad campaigns and spend, store orders and products). Fivetran keeps the sign-ins to them; we keep only daily totals |
| Amazon Web Services (RDS Postgres) (being added) | The reporting store Fivetran lands each client's ad and store data in (a Postgres database in our own AWS account), read back as daily totals | The rows Fivetran brings in from the connected apps, kept apart per client; we open only the asking business's own database, read-only |
| Composio (being added) | One-click app connections: Digital Minions' own Composio project holds each business's connected apps (one user id per business), so the client does not need a Composio account | What the client authorizes for each app it connects. The app sign-ins stay at Composio; we keep only the connection |
| Cloudflare (call relay) | Relays the encrypted audio and video of calls between people inside Digital Minions, for networks that block direct connections (a TURN relay) | The encrypted call media in transit, which it cannot read, and the network addresses of the people on the call |
| Vimeo | Reads the caption text of a Vimeo video a client links for a meeting import, through Vimeo's API with Digital Minions' access token | The video's address and ID; its caption text comes back |
| DataForSEO (being added) | SEO data (keywords, rankings, links), through Digital Minions' own DataForSEO account | No customer personal data sent. Only the keywords and web addresses being researched |
| Keywords Everywhere (being added) | SEO data (keyword volumes and related keywords), through Digital Minions' own Keywords Everywhere account | No customer personal data sent. Only the keywords being researched |
| Google Places | Reads the public review total and star rating of the business's own Google listing once a day (new Google reviews), with Digital Minions' Google key, or the business's own if it adds one | The business name and address (to find the listing once), then the listing's public review total and rating. No customer data |
Your own accounts (you connect them)
These are your own accounts. Our app reads or sends there only with the access you give it, and the company is your provider, under your terms with it.
| Company | What it does for us | What it receives |
|---|---|---|
| Google Workspace (Calendar, Gmail, Drive, Analytics, Search Console, Business Profile, Ads) | The client's Google services, through our Google app | What a minion reads or sends there, with the access the client gives |
| Microsoft (365, Teams, Entra) | The client's mail, calendar, files and Teams; sign-in | What a minion reads or sends there, with the access the client gives |
| Slack | Minions in the client's Slack; Sign in with Slack | Messages to and from minions in the channels they are in |
| HubSpot, QuickBooks, Zoom | The business apps the client connects | What a minion reads there, with the access the client gives |
| Other connected tools (the client's own MCP servers, recipes and notetakers) | Tools the client connects with its own account or key | What the client authorizes |
| The client's own AI keys | When an owner adds their own Anthropic, OpenAI, xAI or Gemini key, that provider bills them directly | As for that AI provider above |
Checked and not used
- Cloudflare: Used only for the call relay listed above, nothing else. A client may point backups at Cloudflare R2 instead of Amazon S3, and then R2 is that store.
- GitHub: Holds our code and runs our tests. No client data goes there.
- Analytics and advertising trackers: None in the app.
AI providers and training
Each AI provider works under its own terms.
- Anthropic: its Commercial Terms say Anthropic may not train models on customer content from its services.
-
OpenAI: it says data sent to its API isn't used to train its models unless you opt in. We send
store: false. -
xAI: its API FAQ says it doesn't train on API inputs or outputs without your explicit permission. It keeps requests and responses for 30 days, for auditing in case of suspected abuse or misuse. We send
store: false. - Google: it says that on the paid Gemini API it doesn't use prompts or responses to improve its products. The free tier does.
Where they run
Your workspace and its backups stay in the region chosen when it was made.