Bernstein

Bernstein is an open-source Python tool for running multiple AI agents in parallel under rules set in configuration files. The rules specify which roles may perform which tasks, which actions require approval, and what is logged. The tool was developed for command-line coding agents such as Claude Code, Codex and Gemini CLI, and supports more than 50 of them. It can also run tasks that produce reports, datasets or audit documents.

The user gives Bernstein a goal, which a language model splits into tasks, each assigned a role such as architect, security or quality assurance. A scheduler written in Python then assigns the tasks to agents, which work in parallel in separate workspaces. Rules set in configuration files define which roles may carry out which tasks and where human approval is required. Before any output is accepted, automated checks confirm defined conditions, for example that tests pass or that required files exist. Output that fails is retried or passed to a different model.

Each run is logged in a journal that can be replayed to compare results step by step. An optional audit log links entries using cryptographic hashes, so later changes to the record can be detected. A run can also be exported as a digitally signed receipt that a reviewer can check offline. The developer notes that some features, including cloud execution, are still experimental. The code is released under the Apache 2.0 licence.