Codev is a novel software development operating system designed for human-agent co-development, transforming natural language specifications into tested, reviewed pull requests. It targets engineers and power users seeking to leverage AI for accelerated and higher-quality software delivery, enabling teams to ship features rapidly, with a median feature implementation time of 57 minutes.

How It Works

Codev employs the SPIR protocol (Specify, Plan, Implement, Review), treating natural language context as executable code. Developers write specifications, which an "Architect" (human or AI) refines. Autonomous AI "Builders" then implement these specs in isolated worktrees, generating code, tests, and plans. This structured, document-driven approach, combined with multi-model AI consultation (e.g., Claude, Gemini, Codex), significantly enhances code quality, test coverage, and deployment readiness compared to unstructured AI coding methods.

Quick Start & Requirements

• Install: npm install -g @cluesmith/codev
• Prerequisites: Node.js 18+, Git 2.5+. Recommended AI CLIs: @anthropic-ai/claude-code, gemini-cli, @openai/codex. Optional for Agent Farm: gh (GitHub CLI).
• Setup: Initialize with codev init, then spawn a builder for a GitHub issue using af spawn <issue-number>.
• Links: FAQ, Tips & Tricks, Discord, CLI Reference.

Highlighted Details

• Performance: A single architect with autonomous AI builders matched the output of a 3-4 person elite engineering team, shipping 106 PRs in 14 days. SPIR consistently shows a significant quality advantage over unstructured AI coding, particularly in test coverage and deployment readiness.
• AI-Native Workflow: Supports multi-agent consultation, where different AI models contribute specialized strengths (e.g., edge-case analysis, runtime semantics, architecture) to catch bugs and improve code.
• Self-Hosting: Codev is developed using its own methodology ("eating our own dog food"), ensuring the framework is battle-tested and evolves based on real-world usage.
• Agent Farm: An optional dashboard for managing and monitoring multiple AI agents working in parallel on development tasks.

Maintenance & Community

Codev is actively maintained through its self-hosting development process. Community engagement is fostered via GitHub Discussions and a dedicated Discord server.

Licensing & Compatibility

The project is released under the permissive MIT License, allowing for broad adoption and integration into commercial and closed-source projects without significant restrictions.

Limitations & Caveats

The Agent Farm component is currently optimized for Claude Code and is primarily macOS-focused, though it may function on Linux. Enabling autonomous AI builders requires using flags like --dangerously-skip-permissions or --yolo, which bypass confirmation prompts and necessitate a high degree of trust in the AI's actions within development environments.