Bernstein is a deterministic orchestrator for CLI coding agents that coordinates parallel execution of AI-driven code generation tools without placing a language model in the coordination loop itself.
The problem Bernstein addresses is the need to run multiple CLI coding agents in parallel while maintaining reproducibility and auditability. Its approach isolates each task in separate git worktrees, allowing parallel runs to replay byte-identically. This deterministic execution model removes the model from the orchestration layer, replacing it with a coordination system that can be audited and verified. The tool provides signed lineage tracking and an optional HMAC audit chain that reviewers can verify offline, creating a tamper-evident record of how code was generated and modified.
Bernstein suits teams deploying autonomous coding agents at scale who need reproducible results and strong auditability guarantees. It works with over forty CLI coding agents including Claude Code, Codex, and Gemini CLI. The project supports both cluster mode for distributed execution and air-gap deployment for isolated environments, making it applicable to security-conscious organizations that cannot route agent execution through external services. Teams should adopt this tool if deterministic replay and offline audit verification are requirements; those prioritizing simplicity over reproducibility may find the additional infrastructure overhead unnecessary.
The project maintains active development with regular commits addressing core functionality and bug fixes. Work focuses on expanding agent compatibility and refining the deterministic execution model. The maintainers respond to issues and pull requests, indicating ongoing engagement with users. Development activity shows attention to the orchestration layer and audit mechanisms that distinguish the tool, with less frequent but meaningful updates to documentation and deployment examples.
