Bernstein

•
1.2k
•

Orchestrates over 30 CLI AI coding agents to deterministically build software, run tests, and merge working code.

Bernstein is a deterministic orchestrator designed to streamline the software development process using multiple AI coding agents. Unlike LLM-driven orchestrators that can be non-deterministic and costly, Bernstein uses a single LLM call for initial task decomposition, then relies on Python for parallel agent execution, git worktree isolation, robust verification through quality gates, and automated merging. It supports over 30 CLI AI coding agents, offering flexibility without vendor lock-in, and ensures every run is reproducible, auditable, and traceable, allowing developers to focus on higher-level goals while Bernstein handles the code generation and integration.

Key Features

01Isolated git worktrees and pluggable sandboxes (Docker, E2B, Modal) for secure agent execution.
02195 GitHub stars
03Built-in Model Context Protocol (MCP) server mode and A2A protocol support.
04Integrated janitor verification with quality gates (tests, lint, types, PII scan) for code quality.
05HMAC-chained audit logs and WAL-backed crash recovery for full traceability and reliability.
06Deterministic orchestration for 30+ CLI AI coding agents with reproducible runs.

Use Cases

01Orchestrating multi-stage coding projects using declarative YAML plans for complex workflows.
02Automating software development tasks, from feature implementation to testing and merging.
03Integrating AI agent-driven development into CI/CD pipelines for autonomous code generation and verification.