GitHub - ygwyg/okiro: Spawn ephemeral, parallel versions of your codebase to ship faster with AI.

2 min read Original article ↗

okiro

Spawn parallel AI coding variations. Compare results. Promote the best.

What is this?

AI agents don't always get it right the first time. okiro lets you run multiple agents on the same task, see their different approaches, and pick the winner.

okiro 3 --prompt "add dark mode" --run
# Creates 3 variations, launches AI agents in each
# Compare results, promote the best one

Install

Requires Node.js 18+.

Quick Start

# Create 3 variations and run AI agents
okiro 3 --prompt "add user auth" --run

# Compare changes visually
okiro compare

# Let AI judge the variations
okiro judge

# Promote the winner
okiro promote var-1

Commands

Command Description
okiro <n> Create n variations
okiro status Show active variations
okiro compare Open visual diff viewer
okiro judge AI-powered ranking
okiro diff var-1 CLI diff vs original
okiro promote var-1 Apply changes to original
okiro cleanup Remove all variations

Options

Creating Variations

okiro 3                              # Basic: create 3 copies
okiro 3 --prompt                     # Prompt for direction per variation
okiro 3 --prompt "add auth"          # Set base task + per-variation directions
okiro 3 --prompt "add auth" --run    # Auto-run AI agents

AI Agent Options

--run              # Auto-detect: claude > opencode > codex
--run=claude       # Force Claude Code
--run=opencode     # Force OpenCode
--run=codex        # Force Codex
--model opus       # Use specific model for all
--model            # Prompt for model per variation

Other Options

-f, --force        # Replace existing variations
--no-terminal      # Don't open terminal sessions
-c, --commit       # Git commit after promote

How It Works

  1. Zero-cost clones - Uses APFS (macOS) or btrfs (Linux) for instant copies that share disk space until modified.

  2. Isolated workspaces - Each variation lives at ~/.okiro/<project>/var-N/. Your original is never touched.

  3. Smart diffing - Ignores node_modules, .git, dist, etc.

  4. AI-powered judging - Analyzes each file across variations and synthesizes a final ranking.

Example Workflow

cd my-app
okiro 3 --prompt "add user auth" --run

# Terminals open with AI agents working...
# When done:

okiro compare          # Visual comparison
okiro judge            # AI ranking

okiro promote var-2 -c "feat: add auth"
okiro cleanup

Docs

License

MIT