Troubleshooting¶
Rate limit errors¶
Increase the per-hour cap:
Or disable limits entirely for a single run:
See Rate limiting for the full syntax.
No models in the dropdown¶
You haven't set a provider API key that matches any enabled entry in config/providers.yaml. Set at least ANTHROPIC_API_KEY in .env, or edit the provider registry.
The container logs this on startup when no providers are enabled:
Permission errors on the database¶
The container runs as a non-root user (appuser, UID 1000). The host ./data directory must be writable by that UID.
Template not found¶
Templates must live at templates/ in the project root, not inside src/templates/. This applies if you're building your own image or running locally without Docker.
Container uses a lot of memory¶
~300MB resident is normal for the Bakwas container. The Python runtime, LiteLLM, and yt-dlp together account for most of it. If you need a smaller footprint, drop Gunicorn to a single worker by editing Dockerfile:
That saves roughly 100MB with a small tradeoff in request parallelism.
Long LLM calls time out behind Cloudflare¶
Cloudflare's proxy enforces a 100-second response limit. Regenerating a long video through a slow model can sometimes exceed it. Either switch to a faster model for that request, or set the DNS record for the Bakwas subdomain to DNS-only (grey cloud) to bypass the proxy.