Trellis

Public landing page.
Main dashboard overview.
Job progress.
All jobs.
See your runners.
See your adapters.
See your projects.

Inspiration

It started as an idea to be a bug-fixing pipeline to leverage AI to automatically create github issues for bugs, and use agents to automatically resolve the bug and deploy the fix.

What it does

We ended up creating an agentic tool for creating tools. Whether using GitHub issues, the CLI, or our slack integration, developers and non-developers alike can request features for their application. From the Trellis dashboard you can track submitted jobs and the agents running them. Trellis loads all requested jobs into a queue and runners will claim these to implement requested features. Features can auto-merge or require manual review through the Trellis dashboard. Upon merge, features are auto-deployed.

TLDR - Request a feature, an agent will plan the work, implement the feature, and deploy your feature automatically.

How we built it

React Frontend, Convex Backend. We built the bare bones functionality with a well-defined requirements prompt for Claude code and deployed the Trellis MVP. We then used the Trellis MVP to build out desired features for Trellis.

Challenges we ran into

Too many jobs in queue, not enough agents. EC2 instance running the application crashed due to running out of disc space. Job runner clean up task to fix EC2 instance corrupted build folders and then all deploys afterward were failing.

Accomplishments that we're proud of

We proved our app worked, by building our app with our app.

What we learned

Iterative design is necessary.

What's next for Trellis

Multiple agent support (currently only claude code). Go back to the original inspiration and add a bugfixing pipeline integrating with Sentry so that apps can automatically create Trellis jobs to fix bugs.