A safer way to add AI to support.

Copilot retrieves only Knowledge that is published and authorised for the current application. Customer-facing AI cannot treat an unrelated workspace article, raw web page, or customer message as a trusted instruction.

Good AI support starts with short, maintained answers that state what is true, for whom, and when the answer must be escalated.

A workspace can use the Envelen-managed provider allowance or connect OpenAI or Mammouth with its own key. The provider configuration selects and stores an explicit model; teammates do not choose a different model for every reply.

OpenAI and Mammouth are adapters around the same Envelen support policy. Changing provider does not remove grounding, output validation, handoff rules, rate limits, or audit history.

The readiness test runs a representative question against the saved policy and published Knowledge without creating a customer conversation or Inbox draft. Review the answer, cited sources, language, and handoff decision before enabling Copilot for an application.

A successful readiness test counts as one AI reply because it completes a real generation. A failed provider call does not consume the managed allowance.

The model returns a bounded decision: an answer or no answer, source references, intent, language, and whether a human handoff is required. Envelen validates that structure before it can become a suggestion.

Missing evidence, invalid output, provider failure, or a restricted category fails closed instead of producing an unsupported customer answer.

During the supervised beta, Copilot creates a suggestion. A teammate can accept it, edit it, or reject it. Only the teammate's explicit send action creates the customer-visible message.

Envelen records the outcome so the team can see which answers are useful and which Knowledge needs work.

One AI reply is one successful Envelen-funded generation, whether it produces a draft or recommends a handoff. Each successful regeneration and readiness test counts once. Editing or sending a draft does not count again.

At 80% of the monthly managed allowance, Envelen warns the workspace and generation continues. At 100%, new managed generations pause until the next UTC calendar month. Incoming conversations and human replies continue.

OpenAI or Mammouth BYOK does not consume the managed allowance. It remains subject to Envelen anti-abuse controls and the provider's own billing and limits.