EgisAI

EgisAI is a runtime control plane for AI agents. It sits between an AI system and the real world to block destructive tool calls, mask PII, and keep an audit of every action. The product is described as working across OpenAI, Anthropic, Gemini, Bedrock, LangChain, CrewAI, AutoGen, and 12+ more frameworks, with one line of code to initialize it.

Its runtime features include automatic agent identity based on system-prompt fingerprinting, so agents appear without manual registration or agent IDs. EgisAI also records behavioral fingerprints such as cadence, model affinity, tool-call signatures, and bucketed prompt-shape histograms per agent and per end-user. It performs Z-score-based anomaly detection on every dimension, grades findings by severity, and generates a per-agent trust score built from provenance, cadence, anomaly density, and policy alignment. The policy layer uses local deterministic checks first for PII, regex patterns, model allow-lists, and prompt-size limits, then invokes an LLM-based semantic guard only when needed. It can also deny tool calls, MCP calls, bash commands, database queries, and financial actions before dispatch. An append-only audit trail with run-level identity stamping supports on-demand SOC 2, ISO 27001, and HIPAA evidence packets.

The page presents EgisAI for teams shipping AI features in production, especially where agents call tools and touch data. It says supported AI libraries are patched in process at import time, and that policy updates ship in seconds via SSE. The system is described as failing open on availability but closed on PII.