Open-Source Tool 'Bartholomew' Offers Sub-35µs In-Process Security Gating for AI Agents
Developers have launched Bartholomew (BTP v3.0) and mcp-proxy-guard, a free, open-source security toolkit designed to protect autonomous AI agents such as Claude Desktop, Cursor, and LangChain from executing harmful actions. The tools address key weaknesses in existing AI guardrails, including slow prompt classifiers that take up to 2,500ms to evaluate actions and unmasked API keys leaking into model context windows. Unlike external chat-based safety filters, Bartholomew operates directly inside the agent's process memory, performing AST syntax tree inspection in under 35 microseconds to block destructive commands before they reach the OS or database. The toolkit also includes in-flight credential redaction and in-memory micro-rollback snapshots to prevent residual damage when a safety rule is triggered. Released under Apache 2.0 and MIT licenses, the project is available on GitHub, PyPI, and npm at no cost.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)
Log in to join the discussion and vote.
Log in