bernstein

Why we included this project

Running several CLI coding agents at once usually means babysitting overlapping edits and output that never quite repeats. Bernstein takes a goal you declare, breaks it into tasks, and runs short-lived agents in parallel, each in its own git worktree, then checks and merges the results. No model sits in the coordination loop, so the same input replays byte-identically; that makes a run reproducible and easy to trust. A signed lineage and an optional HMAC audit chain let a reviewer verify the whole pipeline offline, and cluster mode plus air-gap deployment handle bigger or isolated setups. If you manage a fleet of coding agents and want reproducible, auditable parallel execution instead of another single-agent wrapper, this is worth a look.