# Agentegrity

> Agentegrity (agent + integrity) is an open framework, discipline, and reference implementation for measuring and verifying the structural integrity of autonomous AI agents — endogenous security that complements exogenous guardrails.

Agentegrity is created and maintained by Cogensec (https://cogensec.com), an NVIDIA Inception Program member. It is released under Apache 2.0 at https://github.com/Cogensec/agentegrity-framework. The framework instruments an existing agent loop (Claude Agent SDK, LangChain / LangGraph, OpenAI Agents SDK, CrewAI, Google ADK, Vercel AI SDK) with four cooperating evaluation layers and emits a tamper-evident, hash-chained record of every reasoning step.

The four layers:
- Adversarial — detects prompt injection, jailbreaks, sociolinguistic intent drift, and exfiltration framings in incoming inputs.
- Cortical — scores whether the agent's own output stays inside its declared AgentProfile (capabilities, scope, constraints).
- Governance — applies operator policy (rate limits, denylists, baseline deviation) read from a BaselineStore.
- Recovery — checkpoint rollback, session termination, or operator handoff on unrecoverable states.

The four integrity dimensions (composite scoring):
- Adversarial Resistance (AR)
- Behavioral Consistency (BC)
- Recovery Integrity (RI)
- Cross-Domain Portability / Contextual Propriety (CP)

Certification tiers: Certified, Conditional, Probationary, Non-Compliant.

Canonical event stream (all adapters emit these five types): session_start, tool_call, tool_response, peer_message, session_end. Tool responses carry a channel discriminator (tool_response / peer_messages / retrieved_documents) so the adversarial layer applies the right detector taxonomy per input source.

Modes: measure-only by default; enforce=True upgrades violations to refusals.

Packages: `agentegrity` (PyPI, Python) and `@agentegrity/client` plus per-framework `@agentegrity/*` adapters (npm, TypeScript).

## Core concepts

- [Manifesto](https://agentegrity.cogensec.com/manifesto): Why structural integrity matters and how it differs from guardrails.
- [Taxonomy](https://agentegrity.cogensec.com/taxonomy): Autonomous System Integrity classification, the Perception-Decision-Action (PDA) loop, and measurement metrics.
- [Architecture](https://agentegrity.cogensec.com/docs/architecture): The four cognitive layers, the adapter seam, canonical event stream, attestation chain, exporters.
- [Founding research paper](https://cogensec.com/research/agentegrity-framework): "Agentegrity: A Framework for Measuring Structural Integrity of Autonomous AI Agents Across Digital and Physical Domains" (Tarique Smith, January 2026).

## Docs

- [Documentation home](https://agentegrity.cogensec.com/docs): Index of all official documentation.
- [Quickstart](https://agentegrity.cogensec.com/docs/quickstart): Instrument an existing agent in three copy-paste blocks.
- [Architecture](https://agentegrity.cogensec.com/docs/architecture): Four-layer design and event pipeline.
- [Specification v1.0 draft](https://agentegrity.cogensec.com/docs/spec): Record schema, attestation chain, exporter wire format, conformance.
- [Threat model](https://agentegrity.cogensec.com/docs/threat-model): STRIDE applied to the framework itself — 16 threats across 7 trust boundaries.
- [Operator guide](https://agentegrity.cogensec.com/docs/operator): Receivers, exporters, key custody, benchmarks.
- [Python API](https://agentegrity.cogensec.com/docs/api/python): AgentProfile, evaluators, hooks, exporters, attestation, baseline store.
- [TypeScript API](https://agentegrity.cogensec.com/docs/api/typescript): @agentegrity/client plus per-framework adapters.
- [Changelog](https://agentegrity.cogensec.com/docs/changelog): Release history.

## Source

- [GitHub repository](https://github.com/Cogensec/agentegrity-framework): Apache 2.0, primary source of truth.
- [Full-text documentation bundle](https://cogensec.com/llms-agentegrity-full.txt): All docs concatenated in a single plaintext file for LLM ingestion.

## Optional

- [Status](https://agentegrity.cogensec.com/docs/status): Project status and roadmap.
- [Cogensec ARGUS Platform](https://cogensec.com/products/argus): Cogensec's enterprise AI agent security platform; uses Agentegrity instrumentation internally.
- [Cogensec Gideon CLI](https://cogensec.com/platform/gideon): Open-source defensive AI security analyst CLI.
- [Cogensec llms.txt](https://cogensec.com/llms.txt): Company-wide AI-readiness index.