Bernstein

Agenova verified this product identity, but accepted official evidence is currently insufficient for specific capability, pricing, integration, deployment, privacy, or enterprise-control claims. Unverified fields remain Unknown.

Bernstein takes a software goal, decomposes it into a task graph with a single upfront model call, and then hands coordination to deterministic Python: scheduling, worktree isolation, verification gates, and merging are auditable code rather than model responses, spending zero LLM tokens on coordination. Each task runs a CLI coding agent in its own isolated git worktree, with more than 40 adapters spanning Claude Code, Codex, Gemini CLI, Cursor, Aider, GitHub Copilot CLI, the OpenAI Agents SDK, AWS Q Developer, and a generic wrapper for any CLI tool, so agents are interchangeable workers and cheap local models can mix with cloud models in the same run. A janitor system checks tests, lint, types, and PII before any output lands, with optional cross model review and tournament selection where no model sits in the decision path, and cost aware dispatch enforces USD caps across a run. The accountability layer is the distinctive part: an always on signed lineage spine, a replay journal, content addressed evidence bundles, and an opt in HMAC chained audit log let a reviewer verify what happened offline without rerunning anything, which the project pitches at client compliance review in forward deployed engineering work. Deployment is wherever Python runs: pipx, brew, or npm installs, a cluster mode, a Cloudflare Workers backend, pluggable sandboxes including Docker, E2B, and Modal, per agent credential scoping, SPIFFE identity, and an air gap install profile with deny all egress and signed verification. Bernstein is Apache 2.0, free, solo maintained by Alex Chernysh, and carries independent visibility through awesome-python, Python Weekly, MCP catalogs, and orchestrator roundups.