A local safety layer for AI outputs and agent actions.
Guardian turns shared principles into inspectable decisions before text reaches people or agents take risky actions. Run it locally, integrate it through Python, and inspect every decision path.
Use the app for personal and institutional checks. Use the API to approve, revise, block, or escalate agent actions before they run.
Read the Guardian API notesGuardian benefits
Apache 2.0 Local Mode
Core local code has no built-in Creed Space subscription or usage meter. Operator hardware, models, connectivity, and third-party costs remain separate.
Local-First Privacy
Core safety checks run locally. No account required for local mode.
Works Offline
After one-time setup, core safety checks run offline. Air-gap compatible.
Open Source
Inspect, modify, contribute. Full transparency in how safety decisions are made.
Compiled Safety
Safety rules compile to deterministic code. Near-zero cost, perfect consistency, no LLM needed for structural checks.
Same Technology. Your Choice of Interface.
Same personalised constitutional checks. Same local-first processing boundary for core evaluation. Downloads, updates, sign-in, cloud sync, and external links may use the network.
Personal, research, developer, and institutional paths share one Guardian loop
People using companion AI, parents or guardians, therapist-adjacent reviewers, and wellbeing advocates
Personal Companion Safety
Human-facing safety that supports continuity, negotiated values, care, and inspectability.
- Job
- Express personal boundaries as selected rules, understand why responses changed, and keep appeal paths visible.
- Recommended mode
- balanced
- Trust artifact
- Plain-language record showing input summary, boundary rule, decision, rationale, confidence, and revision path.
Researchers, lab leads, evaluators, and funder reviewers
AI Safety Research
Research tooling for safety evaluation, runtime evidence, and reproducible review.
- Job
- Compare Guardian decisions with no custom rules, one rule, and multiple rules so safety claims can be inspected rather than trusted.
- Recommended mode
- balanced
- Trust artifact
- Research matrix record with rule stack, configuration, outcome, rationale, uncertainty, and integrity hash.
Agent-platform developers, agent team leads, and safety engineers
Guardian Layer for Agents
An integration story for agents that need a decision gate before risky actions.
- Job
- Insert Guardian between agents and risky actions so tool calls, messages, context changes, and policy updates are evaluated before execution.
- Recommended mode
- custom
- Trust artifact
- Machine-readable action record with action hash, rule references, decision token status, and consumer guidance.
Public-interest teams, labs, schools, clinics, nonprofits, and accountable deployers
Institutional AI Safety Governance
Inspectable decisions, review trails, and clear limits for consequential deployments.
- Job
- Turn written rules into testable AI safety checks, run decisions around model or agent outputs, and export audit records for review.
- Recommended mode
- strict
- Trust artifact
- Signed decision record with policy hash, actor context, reasons, appeal route, and exportable JSON.
OpenClaw & Moltbot Users: Start Here
Already running an autonomous agent? We built this for you.
Before you give your agent the keys, give it principles.
Guardian Usage GuideWorks with OpenClaw, Moltbot, Claude Code, and any agent that accepts a system prompt.
New: a Hermes / OpenClaw plugin now enforces your constitution in-process. It injects the active constitution every turn and can gate tool calls through a Creed PDP (advisory or enforce). Agent framework integrations guide
From local devices to servers
Guardian can select a default tier based on available memory where supported. You can also choose explicitly.
Common Questions
What does the local licence cover?
Core local Guardian code is Apache 2.0 licensed. Cloud models and hosted services have deployment-specific account, availability, funding, usage, and charge terms.
Do you collect any data?
Core Guardian evaluations run locally with no cloud logging. Optional account features only sync when you sign in.
App or library: which should I use?
Same technology. Use the app if you just want to check text. Use the library if you're building software.
What about Fleet?
Guardian runs locally on your hardware. Fleet adds hosted centralised governance across your organisation; its account, availability, usage, and cost terms are deployment-specific.
On-device vs Cloud
Guardian (On-device)
- Included: Apache 2.0 licensed local code
- Included: Core evaluations stay on your device
- Included: Works offline
- Included: Optional local JSON audit logs
- Limitation: Local models (less nuanced than frontier)
- Limitation: Manual rule-bundle updates (no live cloud sync)
Creed Space (Cloud)
- Included: Cloud or local models, your choice
- Included: Custom rules for your values and policies
- Included: Nuanced, context-aware decisions
- Included: Audit logging and compliance-report design support. Exact deployment coverage, provider retention, legal interpretation, and independent assurance remain open.
- Included: Fine-grained policy controls
- Included: Fleet management dashboard
- Included: Zero setup, works instantly
Guardian runs on-device for private, everyday checks. Creed Space web deployments add team governance, richer context, and audit-record design support, with cloud or local models. Exact deployment coverage, provider retention, legal interpretation, and independent assurance remain open.
Ready to try it?
Core safety checks run locally on your device.
pip install creed-guardian