Inspiration
As developers, we love the idea of "agentic workflows," but the reality is a mess of terminal tabs. Running 5-10 parallel agents means constant context switching: creating worktrees, tracking which branch is "dirty," and manually syncing across windows. We built OpenSwarm on Rails to turn that chaotic "terminal juggling" into a streamlined, keyboard-first command center.
What it does
The project is for developers who wants to be more productive by run parallel AI agents without having to worry about git interruptions and agent orchestration. Our tool offers Extremely parallelizable workflows with convenient keybinds and a vim-like workflow that keeps you moving faster. Not only do we allow collaboration between developers with a zed-like environment but also with managers that set requirements with our Miro integration. We are completely redifining coding. And lets not forget about our innovative sollutions to tackle merge conflicts and creating structured PRDs from a feature description.
Visualize & Manage Worktrees: A live graph shows the status of every branch (dirty, ahead/behind, or active).
Parallel Agent Control: Launch and monitor multiple agents (Claude, OpenCode, or Shell) in embedded terminals on one screen.
Fast-Path Git Operations: Single-key commands for staging, committing, pushing, and merging without ever typing cd.
PRD-to-Code Pipeline: Seamlessly exports PRDs to Miro AI for collaborative editing and imports them back to drive agent execution.
How we built it
We leaned heavily into the Ruby on Rails ecosystem to create a high-performance, real-time UI.
- Backend: Rails 7 handles the complex Git worktree lifecycles and process management.
- Frontend: A "Zed-inspired" keyboard-first interface. We used Hotwire and Tailwind CSS to ensure navigation is fast and low-friction.
- Integrations: We integrated Miro to allow for a circular workflow: Exporting PRDs for planning, then re-importing them to trigger agent execution.
- Desktop: We wrapped the experience in Electron to provide a dedicated workspace away from the browser tabs.
Challenges we ran into
The biggest hurdle was managing real-time state synchronization. Keeping a visual graph of Git worktrees accurate while agents are actively committing and pushing required a robust background job architecture and a snappy frontend that doesn't lag under heavy I/O.
Accomplishments that we're proud of
Build a cli tool first and use the same tool to create UI wrappers around it.
What we learned
We discovered that the "bottleneck" in AI agent productivity isn't just the LLM speed—it's the developer's overhead in managing the environment. By automating the "plumbing" of Git (staging, merging, pruning), we can actually keep up with 10 agents at once.
Log in or sign up for Devpost to join the conversation.