My goal: Use app gen, without it burning credits on our own openAI keys
Issue: Can’t use app gen
Steps I've taken to troubleshoot: talked to support
Additional info: (Cloud or Self-hosted, Screenshots), cloud, team plan.
Hey, we’re not on the enterprise plan (which allows you to use your own key for AppGen), and therefore the configuration of our openAI key shouldn’t affect if AppGen works, but at the moment, I’m blocked from using AppGen as we’ve blocked expensive Gpt4 and Gpt5 models.
Why does this block me from using AppGen? AppGen should be using Retool’s key, not mines.
Hi @Matthew_Carter, the default model on the Retool AI page (gpt-4o-mini) reflects what model will be used for Retool AI components or queries, but doesn't affect what will be used for Assist. By default Assist uses multiple models from Retool-Managed OpenAI and Anthropic accounts, including GPT4 and GPT5 models. I see from your screenshot that the Retool managed OpenAI is Disabled, which is likely why Assist isn't working. Since you aren't on Enterprise, you are correct that it shouldn't be trying to use your key.
On Enterprise, the screen for picking which model to use looks like this:
It is worth mentioning however, that the only difference with Enterprise in this case is that they can configure using a single provider (OpenAI, Azure OpenAI, Anthropic, or Amazon Bedrock), rather than using both Anthropic and OpenAI. Enterprise customers can not customize which models are used by Assist at this time. This is mentioned in the docs here: Configure your organization for Assist | Retool Docs.
Hey @mike_m thanks for the repssonse - Yes I understand the default model isn’t for assist, no worries there. My goal here is to use assist, it has pretty much been broken for us since launch.
So is it intentional that:
Any customer not on enterprise, who configures thier own API key, is therefore unable to use assist? Because configuring your own key disables the retool key for our instance? Because that is the case for us.
Why does configuring my openAPI key that has model restrictions block assist from working? The only reason I can think, is that you are not using Retool’s key as you claim for assist, but you are using our key. In fact, I know this to be true, because if i unrestrict our key, Assist works again, and suddenly something is making expensive GPT4 and GPT5 calls using our key.
From our perspective, we’re just being told we are doing something wrong, but it’s perfectly reasonable that we want to use an OpenAI key with model restrictions for saftey in Retool. I don’t understand why us configuring our key means assist can’t work - because, as you claim, assist should only use retool-managed keys. So why does a restriction on our key break assist?
Thanks for your patience on this Matthew, I was able to find out some more information on this and I can confirm for you that there is currently a mismatch between the docs and how Assist functions as of now. You're correct that Assist is using your key when you unrestrict it.
Our intention is to ultimately have Assist only use Retool-managed keys for non-enterprise customers, which is why the docs reflect that at the moment. However, due to an inflexibility with how Retool AI resources work, we can't have both a Retool-managed resource configured for Assist and a self-managed resource configured for Agents and other features at the same time. Ideally we would have both, and that is being worked on. I am now tracking that ticket and will update this thread when it is finished.
So, in summary, the intention and goal is that all customers be able to use Assist with Retool keys, and be able to use other AI features with their own keys. The current reality is that you have to either use your self-managed key for Assist, use Retool-managed keys for all AI use cases, or continue as you have been with Assist unable to use your restricted key.
