Skip to content

rizviz/GRITS

Repository files navigation

GRITS

Open lifecycle governance and runtime assurance for AI agents and LLM systems.

GRITS helps teams do two things without switching frameworks:

  1. secure and govern one deployment now
  2. govern agent fleets over time

GRITS is not another abstract AI policy framework. It is an implementation-first package of profiles, baselines, lifecycle rules, runtime signals, and assessment artifacts that teams can actually use.

Who this is for

If you are... You probably care about... GRITS gives you...
running OpenClaw, NemoClaw, or similar agent runtimes hardening quickly, reducing mistakes, seeing clear gaps deployment baselines, remediation playbooks, secure defaults, scan/score/remediate workflow
building agents or LLM apps knowing what controls apply without bureaucracy Agent Profile, LLM App Profile, ownership rules, lifecycle model, practical defaults
leading platform, governance, or security at scale ownership, lifecycle, drift, runtime monitoring, recertification managed-object model, Agent Lifecycle Model, transition gates, runtime signal model, scorecards

What you get in this repo

Adoption Engine track

Built for operators, builders, self-hosters, and small teams who need immediate value.

  • OpenClaw deployment hardening baseline
  • OpenClaw remediation playbook
  • secure default patterns for local dev, home lab, and always-on agents
  • starter tooling to classify, check, and score systems
  • sample profiles and examples

Enterprise Reference track

Built for teams planning to govern many agents over time.

  • Agent Lifecycle Model
  • managed-object model for agents and LLM apps
  • ownership and accountability model
  • transition gates and recertification logic
  • runtime signal model and evidence requirements
  • scorecard model and severity mapping

Start here

Path A — I want to harden an existing agent deployment

Read these first:

  1. QUICKSTART.md
  2. baselines/openclaw-deployment-hardening-baseline.md
  3. playbooks/openclaw-remediation-playbook.md

Then run:

python tools/grits_profile_selector_v2.py examples/sample-agent-record.yaml
python tools/grits_openclaw_checklist_v2.py examples/sample-openclaw-posture.yaml
python tools/grits_scorecard_v2.py examples/sample-openclaw-posture.yaml

Path B — I want to define how agents are governed over time

Read these first:

  1. enterprise/agent-lifecycle-model.md
  2. enterprise/agent-registry-schema.md
  3. signals/runtime-signal-model.md
  4. assessment/scorecard-model.md

Why use GRITS

Outcome Why it matters
Saves time teams do not have to invent lifecycle governance and deployment baselines from scratch
Accelerates development builders know what controls apply without dragging in full enterprise bureaucracy
Promotes trust ownership, lifecycle state, review dates, and runtime signals are made explicit
Improves security hardening baselines and remediation playbooks reduce avoidable mistakes
Reduces drift recertification, runtime events, and transition gates make governance continuous
Scales better one agent and one thousand agents can be governed with the same framework language

Framework comparison, value first

Framework Best used for Where GRITS adds value When GRITS is the faster first move
ATF Zero Trust-style governance for autonomous agents adds deployable baselines, lifecycle artifacts, runtime signals, and adoption-ready implementation guidance when builders or operators need something they can apply immediately to an existing runtime
NIST AI RMF broad enterprise AI risk management alignment translates high-level governance outcomes into deployable controls, baselines, evidence, and scorecards when a team needs an operational package, not just a governance umbrella
OWASP agentic or GenAI guidance risk identification, threat awareness, and mitigation categories organizes those concerns into profiles, lifecycle governance, baselines, runtime signals, and scorecards when a team wants one operational frame instead of separate guidance documents
AAGATE platform-oriented continuous governance and control-plane thinking provides an open framework layer that can be used without adopting a full platform architecture when the goal is lightweight adoption, open artifacts, and runtime-agnostic structure

A longer comparison is in docs/comparisons.md.

The GRITS structure in one view

Layer What it does Primary audience
Core defines terms, object model, lifecycle logic, control families, and conformance concepts everyone
Profiles tells users what applies to their system type builders and reviewers
Baselines and playbooks shows how to deploy and remediate in real runtimes operators and self-hosters
Runtime signals defines continuous governance evidence and event vocabulary platform and governance teams
Assessment turns evidence into findings and scorecards security, governance, operators

Immediate release logic

GRITS ships both tracks together:

  • Adoption Engine proves GRITS is useful today
  • Enterprise Reference proves GRITS can scale beyond one host or one agent

If GRITS shipped only baselines, it would stay a hardening pack. If GRITS shipped only enterprise references, it would be read and ignored.

Repository map

core/         framework definitions and conformance concepts
profiles/     Agent and LLM App profiles
baselines/    deployable minimum baselines
playbooks/    remediation and operations playbooks
signals/      runtime signal and evidence model
assessment/   scorecards, findings, severity, remediation mapping
enterprise/   lifecycle, registry, ownership, transition, runtime assurance refs
tools/        starter scripts
examples/     sample records, posture files, events, and scorecards
docs/         audience-facing comparison, metadata, FAQ, roadmap

Release boundaries

Included now:

  • lifecycle model
  • managed-object model
  • Agent Profile and LLM App Profile
  • OpenClaw baseline and remediation playbook
  • runtime signal model
  • scorecard model
  • starter scripts and examples

Deferred beyond this starter package:

  • full control-plane product design
  • sector overlays
  • deep agent-to-agent delegation trust model
  • automated multi-runtime discovery
  • certification program

Contributing

See CONTRIBUTING.md. The most valuable contributions are:

  • runtime-specific baselines
  • lifecycle edge cases
  • signal and event schema feedback
  • example scorecards and real remediation sequences

About

GRITS is an open implementation framework for governing and hardening AI agents and LLM applications.

Resources

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages