Bring Your Own AI Model to QuestDesk: Keys and Connections

You should be able to understand a business tool before you give it a model key. QuestDesk starts with that principle: answer a few setup questions, see a proposed goal and first quests, and decide whether the plan is useful, all before you connect AI. Setup opens inside QuestDesk once you have access; joining the waitlist takes only your email.

A connected model becomes useful later, when a quest has a brief you approve and a result you want to inspect. This guide explains the difference between a subscription and an API key, what QuestDesk can do in early access, and the decisions that stay with you.

A subscription is not an API connection

Paying for ChatGPT, Claude or another chat app gives you access to that product. It does not automatically give a different app permission to call the provider's API, read your chats, or use the subscription's usage allowance. An API key is a separate credential issued for programmatic calls, with its own billing and limits.

If you choose a provider connection, check its pricing and cap settings in that provider account. QuestDesk should show its own known usage honestly, but a run count is not a provider invoice. If a cost is not available from a source, the honest value is unknown.

What QuestDesk asks you to connect

The invite-only app supports a cloud model connection on your own key, with daily run and token allowances per business. The available providers and setup steps are shown when you choose a connection. Other worker routes, including local or code-focused tools, are still being built and tested. An existing external chat is not automatically a QuestDesk worker.

The 3D Kingdom on the public homepage is a product preview. The roadmap distinguishes what invited members can use today from parts still in development. Do not treat an animated worker in that preview as evidence that a paid provider is connected or running.

A useful connection flow should tell you:

• Which provider receives the brief and what account pays for usage.
• What the worker is allowed to return: a draft, a question, or another explicitly supported result.
• What daily allowance applies and what remains unknown.
• How to disconnect the worker without losing your quest history.

Approval belongs to the actual brief

A model connection is permission to make a particular type of request; it is not permission to start every proposed task. Write the expected output and done check first. Read the brief. Approve that version, then press Start. Editing the brief should send it back for review. A worker question should come back to you rather than being answered by a guess.

The work chapter shows this sequence. For the reasoning behind these gates, read human-in-the-loop AI agents.

Start with one useful result

Choose a quest small enough to judge in one sitting. For example: "Draft three customer interview questions about how local shops get repeat bookings. Do not contact anyone." The result is a draft. You can accept it, request changes, or reject it. No message should be sent on your behalf merely because you connected a model.

If the first draft is useful, you have learned something about both the worker and the brief. If it is not, you still have a record of what was asked and what came back. That is more valuable than an unbounded agent run whose actions you cannot reconstruct.

Inside QuestDesk you see your first proposed quests before you connect anything, so you can connect AI later, once you know what you want it to help with. Join the waitlist to get access.