Bernstein
VerifiedOrchestrates CLI coding agents in parallel with deterministic scheduling and signed audits.
What is Bernstein?
Bernstein functions as a deterministic scheduler written in Python that launches crews of CLI coding agents such as Claude Code, Codex, and Gemini CLI. Each agent operates inside its own git worktree, and the system records every scheduling choice, file change, and cost in a tamper-evident log using HMAC signatures and per-artifact lineage tracking.
Coordination relies entirely on plain Python logic rather than additional model calls, allowing users to replay past runs exactly. Security features include bearer-token authentication for the task server, signed agent cards with Ed25519 keys, and optional zero-trust JWT tokens per session.
The project serves engineering teams that run three or more agents concurrently and operators who require compliance-grade audit trails without external SaaS dependencies.
What you can build with Bernstein
Parallel agent execution
Teams can launch multiple coding agents at once, each in a separate worktree, while a merge queue prevents conflicts and race conditions during integration.
Compliance-sensitive automation
Operators obtain plaintext routing decisions and HMAC-signed logs that remain fully local and verifiable without third-party services.
Cost-controlled experimentation
Users track per-artifact lineage including model, prompt hash, and spend, helping control expenses when testing agent combinations.
Install Bernstein
pipx install bernsteinpipx install bernstein
bernstein init
bernstein run -g "fix the failing test in tests/test_foo.py"- 1Install the package with pipx install bernstein.
- 2Run bernstein init to set up the project structure.
- 3List available integrations using bernstein integrations list --installed.
- 4Execute a goal with bernstein run -g followed by your task description.
- 5Verify results and lineage with built-in CLI commands after the run completes.
Bernstein: pros & cons
Pros
- +Supports 44 CLI agent adapters out of the box with consistent worktree isolation.
- +Provides reproducible scheduling and full audit trails without external dependencies.
- +Includes strong security primitives such as signed cards and per-session tokens.
- +Remains lightweight and fully open source under Apache 2.0.
Cons
- –Requires Python 3.12+ and comfort with CLI tooling.
- –Currently maintained by a single developer.
- –Limited to command-line agents and git-based workflows.
Frequently asked questions
It includes adapters for 44 CLI tools including Claude Code, Codex, Gemini CLI, plus generic prompt wrappers and delegators.
User reviews
Verified reviews from the community shape this listing's rating.
Loading reviews…
Similar agents
Other general-purpose options worth comparing.
