spawn

Architecture Diagram
Architecture Brainstorm

Inspiration

As students, we often find great open-source projects but get stuck before we can even contribute. Large repos can be hard to run, dependencies are messy, and setting up the right environment takes more time than actually learning from the code.

We built this to remove that barrier. Instead of spending hours configuring tools, students can spin up clean environments, test ideas quickly, and run experiments safely, like “what if this project used X instead of Y?” It turns curiosity into action, so learning and contributing happen faster.

What it does

Our app monitors you GitHub repo for any issues. Once an issue is created, it creates a new virtual environment to make the changes and test the code. After tests are done, a PR is opened and the virtual environment is spun down.

How we built it

We used Vultr to manage all our cloud services:

Cloud Compute to create the virtual machines to clone and iterate on the code. We created a singular template which was used as a snapshot. Compute instance was also spun up to host a HashiCorp Vault to store the VM credentials for use by our other apps.
Startup Script runs automatically when a compute instance is created to setup the virtual machine
SSH Keys are generated for every instance created on Vultr. These can be used rather than password based credentials and are stored in HashiCorp vault
Object Storage to store the Terraform state for all the GitHub actions. Without this, the state would be lost and entities wouldn't be possible to manage easily.
PostgresQL used to store relations between issues, VMs used to iterate on it, and the pull requests created and logs
Valkey key-value store used to store logs
Serverless Interface used to call AI models for OpenCode to call and use
VPC used to connect Vultr instances to each other and block external connnections.
GitHub Actions uses Terraform to do the initial setup and manage the VMs. Enables easier scaling without needing to clone the repo and configure a file.
The GO backend monitors the GitHub repo for any issues. Once an issue is created, it calls the HashiCorp vault to get credentials of VMs or calls GitHub actions to create more VMs. It then SSH's into the vault and runs the OpenCode instance
React frontend uses web sockets to provide insight into the current status of VMs, actions, and logs. Can get a preview of the VMs and what they're doing

Challenges we ran into

It takes a while to start up VMs, so it's hard to debug the GitHub actions and Terraform deploys
Figuring out how HashiCorp vault

Accomplishments that we're proud of

Can easily update and spin up/down new resources using Terraform and GitHub actions
Terraform state is persisted using Vultr Object Storage so it can be easily updated
Can take down entire Vultr deployment if needed with a GitHub Action