Deterministic enforcement at the execution boundary.
Waveframe turns human governance into deterministic execution control. Authority is compiled, proposals are normalized, CRI-CORE decides admissibility, and Waveframe Guard prevents blocked actions from reaching production systems.
Human governance becomes production control.
This is the canonical Waveframe architecture: authority is authored, compiled, bound to proposals, evaluated, enforced, and then allowed to touch production only when the contract permits it.
Most AI governance arrives too late.
Logs, dashboards, and prompt instructions can be useful, but they do not reliably stop a bad action at the moment it is about to execute.
- Monitor after execution
- Rely on prompts
- Depend on model behavior
- Cannot prevent execution
Two lines to govern execution.
Install the SDK, bind a published contract, and wrap the function that should never run unless governance requirements are satisfied.
Install
pip install waveframe-guardGuard can enforce locally against published contract artifacts generated by the governance workflow.
SDK Example
from pathlib import Path
from waveframe_guard import install_guard, guard
install_guard(
actor={"id": "user-1", "role": "intern"},
contract_path=Path("contracts/finance-policy-1.0.0.contract.json")
)
@guard
def transfer_funds(amount):
return bank.transfer(amount)Local Guard and Cloud have different jobs.
Local enforcement gives developers fast feedback and resilient runtime checks. Waveframe Cloud is the authority and evidence plane: distribution, registry, audit durability, and receipts.
Local Guard
SDK- Simulation
- Developer testing
- Local policy evaluation
- Cached contract enforcement
- Blocks invalid function execution
Waveframe Cloud
Authority- Authoritative contract registry
- Authority distribution
- Immutable audit logs
- Organizational identities
- Durable receipts
- Production governance evidence
A deterministic governance ecosystem.
Each component has a narrow responsibility, so authority can be inspected, replayed, distributed, normalized, evaluated, and audited without relying on model behavior.