ContextDuty: open-source local-first firewall stops secrets and PII from reaching AI coding tools

Interesting_Big_8974 · reddit · 2026-11-21

ContextDuty is a new MIT-licensed, local-first security layer that prevents secrets and PII from leaking into AI coding assistants, MCP servers and LLM APIs before code is ever committed. Features include Microsoft Presidio-based PII/secret detection, AI-tool-specific IDE ignore files, Git pre-commit and CI/CD scanning, a local HTTPS proxy for AI API traffic, an MCP server that sanitizes tool results before they enter model context, and policy-based blocking, warning and redaction. The author is soliciting feedback from AI infra, MCP, agent-framework and DevSecOps practitioners.

Original post →

More from coding & agent

coding & agent channel →