Skip to content
Operon AI
Legal

Subprocessors

Last updated October 28, 2026

These are the companies that process customer data for us to run Operon, what each one does, and where. We tell workspace owners by email at least 30 days before we add or replace one. Questions go to legal@aioperon.com.

Hosting

Where the Operon app and its data live.

  • Railway · runs the Operon app and its Postgres database, which hold every workspace's records · United States, US West region (California)

Calls

Every live call runs on Vapi. Vapi answers the workspace's number, turns the caller's speech into text, passes the conversation to the AI model, speaks the agent's replies in one of Vapi's own voices, records the call unless the workspace has switched recording off, and writes a short summary when the call ends.

Vapi keeps its copy of each call, recording and transcript included, until Operon deletes it: when the workspace's retention limit passes, or when the workspace deletes its history or the workspace itself.

  • Vapi · carries live calls: the phone line, speech to text, the model, the voice, recordings, transcripts, and call summaries · United States
  • Deepgram · turns the caller's speech into text on every call, as part of Vapi's service · United States

Phone numbers and texts

Operon holds the phone numbers it gives workspaces on its own Twilio account.

  • Twilio · holds workspace phone numbers, passes their calls to Vapi, and carries every text message in and out, including booking confirmations and reminders · United States

AI models

An AI model writes the agent's replies. Every agent runs on Claude Haiku 5.5, by Anthropic, unless Operon sets a different model for the platform or a workspace sets its agent to one. Operon can also route a conversation to the other providers below, either because an agent is set to one of their models or as a fallback when the first model fails or does not answer in time.

On a call, Vapi sends the conversation to the model provider, directly or through Operon. Replies to text messages go from Operon to the provider.

DeepSeek serves its models from China. It receives no data unless Operon switches it on for the platform.

  • Anthropic · Claude models, the default for every agent · United States
  • OpenAI · GPT models, when a conversation is routed to them · United States
  • Google · Gemini models, when a conversation is routed to them · United States
  • xAI · Grok models, when a conversation is routed to them · United States
  • Groq · open-weight models on Groq's hardware, when a conversation is routed to them · United States
  • DeepSeek · DeepSeek models, only once Operon switches them on · China

Email

Mail Operon sends on its own behalf.

  • Resend · sends password resets and the notices a workspace's team asks for about messages, bookings, and callers who need a person, which can include a caller's name, number, and message · United States

Billing

Payment for Operon itself.

  • Stripe · bills Operon subscriptions and holds the payment details customers give at checkout · United States

Bank connections

A workspace can connect its business bank accounts and cards for a savings audit. Plaid makes the connection: the account owner signs in to their bank through Plaid, and Plaid gives Operon read-only access to the account's transactions and recurring charges. Operon never receives the bank sign-in and cannot move money.

  • Plaid · connects the bank accounts and cards a workspace chooses for a savings audit, and passes Operon their transactions and recurring charges, read-only · United States

Integrations you connect

Google Calendar, Microsoft Outlook, Calendly, HubSpot, and Slack, and the webhooks and API keys a workspace sets up, for Zapier for instance, are not Operon's subprocessors. A workspace connects them itself, and Operon sends each one only what that workspace chose to send it, under the workspace's own agreement with that provider.