I built DevSwarm to solve a problem I faced while building multi-agent coding workflows: Git branches are linear, but AI agents are concurrent.
When you have multiple agents (or a human + an agent) working on the same feature, they often step on each other's toes—locking the index, creating merge conflicts, or polluting the working directory.
DevSwarm is an infrastructure tool (CLI) that virtualizes a logical git branch into multiple Nodes .
Each Node gets:
1. An isolated Git Worktree (file system isolation).
2. A dedicated Tmux Session (runtime isolation).
3. A managed Shadow Branch (version control isolation).
4. Auto-generated VS Code / Trae Workspace config for multi-root editing.
This means you can have:
- Node A : You, coding the feature logic.
- Node B : An AI agent, writing tests in the background.
- Node C : Another agent, doing a code review or refactoring.
All running in parallel on the same machine, same repo, same logical feature, but completely sandboxed.
It's written in Go. Would love to hear your feedback, especially if you're building agentic coding tools!
Hey HN,
I built DevSwarm to solve a problem I faced while building multi-agent coding workflows: Git branches are linear, but AI agents are concurrent.
When you have multiple agents (or a human + an agent) working on the same feature, they often step on each other's toes—locking the index, creating merge conflicts, or polluting the working directory.
DevSwarm is an infrastructure tool (CLI) that virtualizes a logical git branch into multiple Nodes .
Each Node gets:
1. An isolated Git Worktree (file system isolation). 2. A dedicated Tmux Session (runtime isolation). 3. A managed Shadow Branch (version control isolation). 4. Auto-generated VS Code / Trae Workspace config for multi-root editing. This means you can have:
- Node A : You, coding the feature logic. - Node B : An AI agent, writing tests in the background. - Node C : Another agent, doing a code review or refactoring. All running in parallel on the same machine, same repo, same logical feature, but completely sandboxed.
It's written in Go. Would love to hear your feedback, especially if you're building agentic coding tools!