Need Urgent Help: Facing Multiple Critical Bugs with My Anna App Build

Developers
1

I’m developing an Anna App that generates Manim animations through an Executa plugin. I have been debugging my application and found that the failure appears to occur before my application code reaches the rendering stage. My app invokes: sampling/createMessage llm.complete to generate Manim code. The request consistently fails with:

HTTP 502 Bad Gateway
anna.partners
Host Error

Cloudflare reports:

Browser: Working
Cloudflare: Working
Host: Error

Evidence that this is not related to my Manim rendering pipeline My render pipeline never starts. The logs show:

RPC METHOD: sampling/createMessage

followed immediately by:

RPCError: HTTP 502

The following stages are never reached:

render_animation
_render_in_background()
manim render
ffmpeg merge

I also added debug logging around the renderer and none of those logs appear because execution fails before rendering begins. Verification using the official example To rule out issues in my codebase, I ran the official example: anna-executa-examples/examples/anna-app-llm-demo at main · whtcjdtc2007/anna-executa-examples · GitHub without modifications. The official demo also fails when calling:

llm.complete

with a simple prompt:

Hello, Anna

and returns:

HTTP 502 Bad Gateway
anna.partners
Host Error

This suggests the issue is reproducible even with the official example. Environment

  • Anna App Developer CLI

  • Anna Runtime Local

  • Executa plugin

  • macOS

  • Reproduced with both:

  • Official LLM Demo

    • Custom Manim application

Question

Is there currently an outage or issue affecting:

llm.complete

sampling/createMessage

Anna-hosted LLM services

or is there a known configuration/API change required for Anna Apps using these endpoints?

Thank you.

2

Hi @dora! :waving_hand: Thanks so much for the detailed write-up and the clean repro with the official LLM demo — that made this super fast to track down. :raising_hands:

Great news: this is on our side, not your app. :tada: Your render pipeline was innocent the whole time — the call was failing upstream at the sampling/createMessage / llm.complete step, well before Manim ever started.

What was happening :magnifying_glass_tilted_left:

The errors you hit (HTTP 502) maybe trace back to our free self-hosted model (Gemma). It’s served on a backend that uses a different internal model identifier than the short name apps see. Our main dashboard chat path translated that name correctly, but the app llm.complete / Executa sampling paths were skipping that translation — so the request reached the model server with a name it didn’t recognize. That’s why the official demo failed too, while paid models (GPT / Claude / Gemini / Qwen, etc.) kept working. :white_check_mark:

The fix :hammer_and_wrench:

We’ve patched both app-facing LLM entry points to resolve the free model correctly, consistent with the dashboard path. Once the next deploy lands, Gemma will work the same from your app as it does in the main chat. :rocket:

What you can do right now :high_voltage:

A couple of quick options to unblock yourself today:

  1. Switch to another available model for now (e.g. a Qwen / Gemini / GPT option in the model picker). Everything except the free Gemma path has been working normally. :blush:
  2. Heads-up on free/self-hosted models: these can occasionally be rotated or retired as we update the catalog. If a model ever stops responding, it’s worth re-opening the model picker and selecting a currently-available one — the list always reflects what’s live. :counterclockwise_arrows_button:

Please give it another try after the next release and let us know how it goes — and if anything’s still off, drop your latest logs here and we’ll jump back in. Excited to see your Manim animation app come to life! :clapper_board::sparkles:

Happy building! :yellow_heart: