caic-xyz docs

Appearance

md ​

Branch-locked development containers for AI coding agents. Each container is locked to a repository-branch pair. Safe parallel work, no conflicts.

The problem ​

AI coding agents work best in YOLO mode (full command execution). But running them locally creates:

  • Branch conflicts: agent changes interfere with your local checkout
  • Test conflicts: simultaneous test runs cause failures and race conditions, leak processes
  • Environment pollution: accumulated dependencies cause hidden bugs
  • Context switching: switching branches while an agent works loses progress

Platform support ​

md is supported on Linux, macOS, and Windows via WSL2.

The solution ​

md gives each agent a complete, isolated container with a full git clone:

  • Run agents on multiple branches simultaneously
  • Switch local branches without affecting running agents
  • Run tests in parallel without conflicts
  • Keep your local checkout clean
  • Delete containers cleanly when done

Quick start ​

bash
# Start container for your current branch (auto-SSH in)
git checkout -b wip origin/main
md start

# Inside the container: run your agent
cd ~/src/<repo-name>
claude

# Back on host: check changes and sync
md diff
md pull

Install ​

bash
curl -fsSL https://caic.xyz/install.sh | bash

Documentation ​

  • Commands: full command reference, workflows (fork, multi-repo, CI)
  • Installation: prerequisites, platform notes, from-source builds
  • Configuration: caches, environment variables, agent configs
  • Harnesses: agent-specific setup (Claude Code, Codex, Gemini, Pi, etc.)
  • Preinstalled Tools: what's available inside every container
  • Tailscale: remote access over your tailnet
  • Architecture: image hierarchy, build pipeline, container lifecycle
  • Design: design rationale, trade-offs, cache injection
  • Security: SSH isolation, capabilities, network restrictions

Getting help ​