AI Briefing
KOSign in

Bernstein – open-source governance layer for AI agents

·2026.10.29 15:24

Key point

The release supports 53 CLI adapters and uses a deterministic Python scheduler to ensure byte-identical replay of agent plans without LLMs in the coordination loop.

Details

  • Bernstein v3.23.0 is an open-source governance layer for AI agents that uses a deterministic Python scheduler to run agents in parallel, ensuring plans replay byte-identically without an LLM in the coordination loop.
  • The system enforces policy as code and produces verifiable records via an HMAC-chained event log and optional Ed25519-signed receipts, allowing third parties to verify run integrity without rerunning the agents.
  • The latest release (v3.14.159) introduced named errors for plan validation, removed five commands with deprecation windows, and added a Muse Code adapter along with Agent Plugins v1.0.0 manifests.
  • The tool supports on-premises, laptop, and Cloudflare Workers environments, with opt-in telemetry that defaults to local-only storage to maintain data privacy.
governanceopen-sourceai-agentsorchestrationaudit-loggingdeterministic-scheduling
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

Hacker NewsNvidia OpenShell: Secure Runtime for Autonomous AI Agent FleetsPyTorchKRMicrosoft and ThinkingBox-Training Released: Qwen3.8-27B Repeat Success Rate Improved 2.8xHacker NewsINIT Labs releases INIT, a Rust agent harness that hot-swaps its own core mid-taskOpenAI BlogGreptile Uses GPT-6 Astra to Automate Application Testing and SecurityRedditHemmingway-1.5: 27B Open-Weight Model Claims 90% Success Rate Against AI DetectorsRedditOpenBonsai open-sources pipeline to compress BF16 models to ~2 GB ternary GGUFs