I am preparing a small web-based physics learning service and would appreciate feedback from people who have deployed similar interactive tools with CyberPanel. The application accepts a typed problem or an uploaded image, sends it to an application service, and returns a step-by-step explanation. A public example of the workflow is this physics ai solver tool.
My current plan is to keep the public frontend and application API behind OpenLiteSpeed, terminate TLS at the panel-managed virtual host, and place uploads in a temporary directory outside the web root. Uploaded images should have strict MIME and size checks, random filenames, short retention, and no execute permission. I also plan to rate-limit both the upload endpoint and the expensive solve endpoint, while using a separate health endpoint that does not call the AI provider.
For caching, static assets and ordinary pages can have long cache headers, but generated answers should not be cached across users. I am also considering separate access and application logs with request IDs so that a failed solve can be traced without storing the original question longer than necessary. Backups would include configuration and application data but exclude temporary uploads and secrets.
Are there CyberPanel-specific settings you would recommend for this pattern? In particular, I would value advice on:
- whether application-level rate limiting or a reverse-proxy rule is easier to maintain;
- sensible OpenLiteSpeed timeout limits for requests that may take 20–40 seconds;
- isolating a worker process from the public web user; and
- avoiding accidental caching of personalised or generated responses.
I am aiming for a modest, secure setup rather than premature scaling. Any examples of reliable vhost, process, or logging arrangements would be useful.