Deterministic safety solutions for probabilistic AI agents

Semantic mistake-memory and guardrail for AI agents. Stops agents from repeating the same failures using causal graphs and semantic matching. Supports MCP, LangGraph, and in-process Python.

Hermes Agent plugin that vetoes tool calls against environments you shouldn't touch — typically production. Declarative YAML policy with glob-based host allow/blocklists, cwd matching, IDN/homograph defenses, and a warn mode for dry-runs.