A ghetto lazy TCP proxy to juggle GPU memory bound apps like Stable Diffusion and Ollama.
Don’t use this in production unless you enjoy pain.

This script exists to make it barely possible to share a single GPU between multiple heavy apps by:

• Listening on a TCP port
• Spawning the target service on first request
• Forwarding traffic once it's up
• Shutting it down after idle timeout

It’s useful when running things like stable-diffusion-webui and ollama on one GPU, but only one at a time.

1. Waits for a request
2. If the target isn't running, it executes a start command and waits
3. Proxies traffic once ready
4. If no traffic hits it for a while, it runs a stop command

Very simple app so just need to build binary:

go build -o revenant

./revenant \
  --listen :8080 \
  --target 127.0.0.1:7860 \
  --start-cmd './start.sh' \
  --stop-cmd './stop.sh' \
  --idle-timeout 10m \
  --start-timeout 30s