App Platform

Hi Everyone, I’m Anmol, a AI developer working with Triple Minds on some personal AI side projects and I’ve been researching deeply about the world of AI chatbots, and I keep coming back to one question that seems to fascinate a lot of developers: How hard is it to build a Candy AI clone from scratch? I’m not trying to copy Candy AI’s branding, but I’m extremely curious about how close an independent developer or small team can get to replicating the functionality of a true Candy AI clone. My Stack for Building a Candy AI Clone Here’s the rough tech stack I’m experimenting with for my Candy AI clone: LLMs: LLaMA-3, Mistral, Mixtral, Falcon Embeddings: bge-base-en-v1.5, text-embedding-3-small (for memory recall in the Candy AI clone) Vector DBs: Pinecone, ChromaDB Backend: FastAPI for serving the Candy AI clone’s inference calls Frontend: Next.js for building the chat UI for the Candy AI clone Voice (optional): ElevenLabs for giving the Candy AI clone a voice Images (optional): Stable Diffusion or SDXL for visuals tied into the Candy AI clone’s dialogue So my big question to the community is: ➡️ Has anyone here successfully built a Candy AI clone entirely with open-source models and custom data? What were the biggest wins or failures in your Candy AI clone journey? Did you fine-tune your models, or stick to prompt engineering for your Candy AI clone? How did you solve memory challenges in your Candy AI clone? Any lessons about integrating visuals or voice into a Candy AI clone without killing performance? I’d love to swap notes, see code samples, or just hear your high-level thoughts on whether a full Candy AI clone is achievable outside of big commercial labs. Thanks in advance for sharing any wisdom about building a Candy AI clone!

It would be great if we could use cloud firewalls to control access to deployed applications using App Platform. I've got some private applications that I would deploy using your PaaS offering immediately if I could firewall it off.

Open the Firewall management tool to allow for traffic moderation of the App Platform.

I would like to see an option to modify Cloudflare behaviour or completely disable it for App Platform, since Cloudflare sometimes acts like a man-in-the-middle and modifies the sourcecode sent by the app, which means you can't really be sure that the client gets what you send from the app.

I'm trying to serve Server-Sent Events (SSE) using Flask on the DO App Platform, but instead of streaming, something on DO's side seems to be caching the entire response and only returning it in one big chunk. I've been trying and trying to debug, and finally found this page which seems to indicate it's not possible? https://www.digitalocean.com/community/questions/how-to-enable-http-response-streaming-in-the-app-platform So I have to migrate off the app platform now I guess? Or am I missing something?

When I create an app on App Platform, DO assigns it a random subdomain at “ ondigitalocean.app ”. I can then define a custom domain for this app too. It would be good be able to block access via the default domain. The use case is that we are using Cloudflare Zero Trust to protect an app, and this works fine except for the fact that the protection can be completely bypassed simply by using the default non-custom domain name.

Runtime logs in the UI is currently on a canvas. So we cannot copy long request IDs and such, which is useful for debugging. So it would be nice if we could just download the runtime logs with a click to search through them easily

As we are currently evaluating a switch from Slack to Discord, we don't want to miss out on the Digital Ocean deploy notifications and alerts from the app platform. Therefore, I wanted to inquire if there are any plans to provide webhooks for this feature in the future or if there are any other known workarounds that would allow us to integrate this into Discord?

FrankenPHP is an exciting modern PHP application server built on top of Caddy. It offers significant performance benefits over traditional setups by running PHP as a long-lived process and integrating directly with Caddy’s robust web server features. I’d love to see native support or official build templates for FrankenPHP + Caddy on App Platform. This would allow Laravel (and other PHP) developers to take full advantage of modern PHP hosting approaches without having to resort to custom VPS setups. Benefits: • Native HTTP/3 support via Caddy • Performance boosts via long-lived PHP processes • Modern deployment stack aligned with other modern languages • Great for Laravel, Symfony, and other modern PHP frameworks Please consider adding support or at least an official guide/template for using FrankenPHP with App Platform.

Now we can deploy a static site from the repository with a Dockerfile inside. But sometimes we have a mono repository with different services and a static website. To start the services, we build one Docker Image of that repository and run the /bin/sh -c "npm run start:**" command for each service. Next, it would be great to use this image for a static website that is already built inside. This feature will help to work with PNPM Workspaces and significantly reduce the time required to build and deploy applications.