🧩 Bernstein: Control Plane for AI Coding Agents

"Automate development like Kubernetes automates containers."

🚀 Overview

Bernstein is a modern control plane for orchestrating AI coding agents—much like how Kubernetes manages containers. Simply define your high-level development goal, and Bernstein will:

1. Decompose the objective into smaller tasks.
2. Assign those tasks in parallel to multiple AI agents running within isolated worktrees.
3. Verify all outputs using a specialized janitor agent that checks code correctness, runs tests, and prevents regressions.

This results in reliable, thoroughly tested code and a clean, linear Git history—without lock-in to any specific agent vendor.

✨ Key Features

• Declarative YAML Orchestration

  • Define complex dev tasks and workflows with simple configs.

• Parallel Agent Execution

  • Run multiple AI agents (e.g., Claude Code, Codex CLI, Gemini CLI) across providers simultaneously.

• Automated Output Verification

  • Every result is checked by a dedicated janitor agent for code quality and test passing.

• Vendor-Agnostic & Portable

  • Prompts, workflows, and commands remain portable between different AI agent providers.

• Continuous Self-Improvement

  • Bernstein can auto-tune agent performance, refine prompts, and evolve routing logic.

💡 Use Cases

• Automate Complex Dev Workflows

  • Implement new features, refactor code, or generate API docs—starting from a single goal description.

• Seamless CI/CD Integration

  • Embed AI-driven code generation, testing, and repairs directly into your pipelines—enable zero-touch, unattended operations.

• Feedback-Driven Optimization

  • Continuously enhance agent coordination and output through automated feedback loops.

🛠️ Example Bernstein Workflow

goal: "Add OAuth2 support to user authentication module"
agents:
  - provider: ClaudeCode
  - provider: CodexCLI
janitor:
  runs: ["pytest", "eslint"]

_{Bernstein brings AI agent orchestration and automation to the heart of your software development process.}