Waveframe Guard

Enterprise AI governance layer for deterministic execution control with policy-bound enforcement and immutable audit tracing.

Waveframe Guard sits at the execution boundary for AI-initiated actions. It builds a governance proposal, routes it through deterministic policy enforcement, and returns a clear decision before your system mutates state.

What it does

• Resolves a stored policy by policy_id
• Builds a proposal from actor, action, and human execution roles
• Enforces deterministic policy checks before execution
• Returns structured outcomes such as allowed, pending, or blocked
• Produces immutable audit records with policy-version traceability

What it does not do

• Execute your business action
• Manage approvals or identity proofing for you
• Replace your system-of-record or workflow engine
• Make post-hoc recommendations instead of enforcement decisions

Install

pip install waveframe-guard

Quick start

from waveframe_guard import WaveframeGuard

guard = WaveframeGuard(
    api_key="wf_test_key_123",
    policy_id="finance-core",
    base_url="http://localhost:8000",
)

decision = guard.execute(
    action={
        "type": "transfer",
        "amount": 5000,
        "system": "finance",
        "resource": "payroll",
    },
    context={
        "responsible": "user-alice",
        "accountable": "user-bob",
        "approved_by": "user-charlie",
    },
    actor="ai-agent-v2",
)

if decision["allowed"]:
    print("Execute downstream action")
else:
    print(decision["status"], decision["reason"])

Decision model

Guard returns deterministic, machine-friendly responses. Typical fields include:

{
  "allowed": false,
  "status": "pending",
  "summary": "AI proposed transfer on finance/payroll",
  "reason": "Approval missing or threshold exceeded",
  "risk_level": "critical"
}

• allowed: whether the action may proceed
• status: allowed, pending, or blocked
• reason: human-readable explanation derived after enforcement
• risk_level: UX-level severity classification for operators

Governance model

Waveframe Guard is designed around deterministic execution control:

• Policies are resolved from stored policy versions, not injected inline at execution time
• Guard may shape proposal structure from contract conditions
• The enforcement kernel determines outcome
• Audit records preserve policy-version linkage and execution trace data

Local development

Run the seeded backend and the example script:

python -m backend.seed
python examples/finance_usage.py

Release status

This repository is being prepared for the v0.2.0 release line.

License

Proprietary. See LICENSE.