Inquir Compute
Deploy apps, databases and background jobs in one place
Start new thread

You built an app with AI. What happened when you tried to deploy it?

by
•

I’m building Inquir Compute, a platform for running apps, databases and background jobs in one workspace.

AI tools are helping more people write working applications. I’m interested in what happens next, when you need to put that app online and keep it running.

If you’ve deployed something you built with AI, where did you get stuck? Build errors, database connections, environment variables, unexpected costs or something else?

Tell me what you built, where you deployed it and what took longer than expected. I’d like to use those examples to decide what to improve in Inquir.

78 views

Add a comment

Replies

Best

2 examples from a Next.js app I self-host with Docker on a VPS behind Cloudflare. One env var gets baked into the bundle at build, so it has to be a Docker build arg, not just a runtime variable. Set only at runtime, the image built and deployed fine, and every uploaded image was blocked in production.

And other one hurt users: Cloudflare cuts a response at roughly two minutes, and my AI calls had a 110 second timeout. A slow generation finished on the server, spent the credit, and the browser got a 524. People retried and paid twice.

 as for the second example, you actually can try Inquir's durable functions, potentially it could help preventing such issues.

check:

 Thanks, durable functions look like the right tool when the work is genuinely long. But for me it wasn't, which is what made it a bug rather than an infra problem.

The fix was small. The OpenAI SDK retries a timed-out call twice by default, and the timeout counts per attempt, so a 60 second limit could quietly turn into three minutes. With SDK retries off and my own timeout at 90 seconds, my code now gives up before Cloudflare does, so the catch block runs and the credit goes back.

Where I can see durable functions helping is batch work, like adding audio to an entire deck. So, how does the client find out the result in that model, polling or push?

How much of the deployment process does Inquir handle automatically when an AI-built app is ready to go live?

 with CLI it can be fully automated, all the features coming to CLI after release ASAP

Database connections were my biggest headache. AI wrote the code, but debugging the setup was all me.

that's a common pain, I'm also trying to sort it out with Inquir, by providing templates and easily deployable DBs. Do you also have problems with migrations or anything like this?

The app worked perfectly on my laptop, then immediately chose violence after deployment.

hmm, common case, because of environment differences, I recommend to test your application locally with docker first, when you're planning to deploy for the first time.

What wa the biggest problem you faced when deploying an ai built app and how did you fix it