AI integration, automation, and advanced engineering solutions within Bubble.io applications.

Need help
1

I’m an AI engineer working extensively with Bubble.io, focusing on integrating advanced AI capabilities into no-code applications.
I specialize in connecting external AI APIs, building intelligent workflows, implementing chatbots, automation systems, and data-driven features within Bubble apps.
I’m interested in discussing best practices, performance optimization, scalable architecture, prompt engineering strategies, and real-world use cases.
I also welcome collaboration, technical discussions, and sharing insights about building production-ready AI-powered solutions using Bubble.

1 Like
2

Here’s what i have learned from my last few years of stacking Bubble:

Offload heavy/extended processing off of Bubble
Besides cost effectiveness, running your own stack lets you work with libraries for specialized tasks. Also:

  1. Bubble’s plugin editor is still not fun to work with
  2. workarounds for running code in a Bubble page (eg. Toolbox plugin) can be tiresome since you’re copy pasting between an IDE and Bubble’s text editor - Bubble needs to expose a code editor as a property input to make this a better experience.

Keep user oauth in Bubble, implement a JWT gateway to link users to your external stack
Trust me, the other way around is not worth the effort. Bubble security works best when you use it’s own privacy rules. You can implement other security protocols like zero trust and WARP if you need more security.

Keep ACID sensitive processes in external stacks (I like Cloudflare’s stack)
No matter what people say and whatever workarounds claim to work, the fact is that Bubble does not expose enough control for proper ACID implementation.

Avoid using the Bubble DB for data that needs to be accessed by AI
Just get Bubble to CRUD to the external DB when needed. Use the right DB stack, don’t shoehorn. I still prefer to keep most data types in Bubble.

Unless you are working with extremely large databases, Cloudflare has a good balance between security, performance and cost between it’s various services.
The best way to work with Cloudflare is to handle routing around your stack with Workers. This keeps connecting Bubble to your stack clean and easy.

For goodness sake don’t use a Bubble DB as an MCP server if you’re not on an enterprise plan
You’re gonna kill the shared servers.

Treat AI as a tool in a stack.
Don’t fall into the hype. You don’t need agents and MCP servers for everything. Keep things boring, boring is good.

  • Pick the right model for the right job. You don’t need frontier models for specialized tasks like sentiment analysis, document OCR etc. Unless it’s cheaper
  • Keep things deterministic as much as possible
  • Use LLM observability tools to keep your LLMs in check

Just sharing.

2 Likes
3

Would love to know more about this. Are you generating a token in a bubble backend workflow?

Wish Bubble would just fix this so we didn’t have to depend on external stacks.

I have used it for some Pages, you’ve cloudflare-pilled me.

2 Likes
4

I use Bubble’s user auth to check that a user has properly logged in > if yes + correct credentials > server-side call my Gateway Worker with the Bubble UUID, timestamp and a secret > check check check > if all good return token.

The logic is that Bubble is already a secure OAuth and I don’t want to build and manage something that is important and can get complex. I just need a gateway to protect my stack.

Though I haven’t fully implemented it since I don’t really have a need for it right now. I was building one for learning and understanding. Most of my endpoints are really light, don’t carry much risk and protected by reverse proxy.

They have good leadership IMO. They’ve made some great investments and partnerships, I like the one with PlanetScale. Hopefully I’ll be able to score a big data project next year and being able to bind PlanetScale to Workers saves a lot of headache.

1 Like