AI & Bring-Your-Own-Key Terms

You connect your own API key from Anthropic, OpenAI, Google or Mistral. Usage is billed to you by that provider under your agreement with them. Outbox Loop does not mark up or resell model usage.

If your key stops working, runs out of quota, or is rate limited, the AI steps in your automations will fail. Each AI step has a failure behaviour you choose — send a fallback message, hand the conversation to a person, or stop.

When an AI step runs, we send the provider the instructions you wrote plus the minimum conversation context that step needs — typically the recent messages and the contact fields you selected. We do not send entire conversation histories by default.

Each provider has its own privacy, retention and training terms. They are not interchangeable, and we show you provider-specific information before you enable one.

Your key is encrypted with a key managed in Google Cloud KMS and is never returned to a browser after you save it. You will only see the last four characters again. Team members who can use AI in an automation cannot read the key.

You can rotate or disconnect a key at any time.

An AI model proposes a reply; it never decides whether a message goes out. Outbox Loop's own rules then check Instagram's messaging window, whether the person opted out, your workspace's limits, and the claims you prohibited. Only then is anything sent.

AI output can still be wrong. Review what your automations say, and use the human handoff for anything that matters — pricing commitments, refunds, legal or medical questions.