Running an open-source AI green screen platform on Swarthmore College Computer Society hardware. 1200 users, 50 community GPUs, one RTX Pro 6000 with 96GB VRAM, and a lot of Docker.
How It Started
CorridorKey Cloud is an open-source tool for uploading green screen footage and getting clean alpha channels back out. James Nye built the app, the neural network inference was designed by Niko over at Corridor Crew, and I containerized it and put it on hardware
The hardware belongs to Swarthmore College Computer Society. We run three servers out of our server room and recently picked up an RTX Pro 6000. I wanted to give back — host the web interface and contribute compute during off hours.
Launch Day Went Wild
CorridorCrew dropped their YouTube video with a link to CorridorKey Cloud in the description. We woke up completely overrun with users requesting accounts.
24 hours in:
- 500 active accounts
- 300 account requests — we used manual approval just to slow things down
- 26 registered GPU nodes
By Day 2, our Discord went from 5 people to 65. James and I set up that server for bug reports and feature requests. We did not expect the community to move that fast.
Day 4 brought our first real security event. We opened account requests back up and got hit with hundreds of bot signups in minutes. I had set up rate limiting and Crowdsec but clearly missed something. Good reminder that no matter how much protection you build, someone will always find a way around it. It is also important ot setup monitoring for things like this. After this event, I completely overhauled the SCCS monitoring stack: deploying Prometheus and Grafana across 8 VMs. Additionally, I added Crowdsec as a guard with our reverse proxy to monitor and real-time block active threats. This was a game-changer for the club.
Where Things Stand Now
Numbers keep climbing. We are at around 1200 active users and 50 GPUs signed up on the platform right now.
The Stack
Every service runs in Docker Swarm across our three servers. Swarm works well here — services distribute across nodes automatically, health checks keep things moving, no separate orchestrator needed.
Traefik sits in front of everything as our reverse proxy.
Crowdsec watches our logs and blocks bad actors at the edge.
Supabase (self-hosted) handles auth and user management for free.
Grafana runs within the Swarm cluster for monitoring.
The storage side handles video uploads and output files onto our NFS share.
We maintain separate Docker Compose configs for development, staging, and production. Each environment has its own networking rules and resource limits.
The Community
What makes CorridorKey Cloud interesting isn't just the tech stack - it's the people behind it. It's a real community-driven compute network, and the 50 GPUs signed up so far come from users who believe in open-source AI tools and want to help make them accessible.
The project lives at github.com/JamesNyeVRGuy/CorridorKey-Cloud. The platform itself is at corridorkey.cloud.
If you've got an idle GPU and want to contribute, or if you just want to try the next generation of green-screening software without buying expensive software, check it out. All the code is open-source and we're always looking for contributors.