Trust · Billing · Certification

Every approved action
leaves a verified trail

SafeClash is the bookkeeper and certifier of the Governed AI OS. It does not make decisions. It records them — with cryptographic precision, Merkle-anchored audit trails, and usage receipts that cannot be disputed.

receipt_id sc_7f3a91b4c2d8e...
proposal_id prop_4a2c1f8b...
action restart_node
outcome executed
approval A = 1
decided_by professor_vasen
usage_atoms 14 atoms · 0.00042 sat
merkle_root a9f3c2b1d4e8...
sealed_at 2026-03-13T09:14:22Z
SafeClash Certified This receipt is immutable. The human approved. The action executed. The record is sealed.
What SafeClash does

Three functions. One purpose.

Trust layer
SafeClash certifies that every action in the system was approved by a human. Not claimed — cryptographically verified. The certificate is the proof.
Billing layer
Usage is measured in atoms — the smallest meaningful unit of computation. Every approved action generates a receipt with an atom count. Payment is the continuity guarantee: π in the formula.
Certification mark
Systems that implement the Governed AI OS protocol can apply for SafeClash certification. The mark means one thing: no action in this system executes without human approval.
Position in the system

SafeClash sits at the end of the canonical loop

OpenClashd owns consent. CLASHD27 owns discovery. Jeeves owns the approval surface. SafeClash owns what comes after: the record, the bill, the proof.

Signal
OpenClashGrid
Govern
OpenClashd
Approve
Jeeves
Certify
SafeClash
Billing model

Usage atoms — not tokens, not requests

A usage atom is the smallest billable unit of governed computation. One atom = one meaningful action that passed through the policy gate. Atoms are counted per proposal, not per API call.

Event Atoms Billed when Includes
Signal observed 0 Never Observation is always free
Proposal created 1 On creation Policy gate, residue calculation
Correlated proposal 3 On creation Multi-node correlation, entropy calculation
Action executed (A=1) 5–20 On execution Bounded execution, receipt generation, Merkle append
Action blocked (A=0) 1 On rejection Rejection receipt, audit entry
Certification issued 100 On issuance Full audit trail verification, certificate signing
MERKLE AUDIT TRAIL — safeclash.com/audit root: a9f3c2b1d4e8f7a2... ├── receipt_batch_2026-03-13 │ ├── sc_7f3a91b4c2d8e... action: restart_node A=1 ✓ │ ├── sc_3b8c2d1a9f4e... action: grid.correlated A=1 ✓ │ └── sc_1a2b3c4d5e6f... action: db.delete A=0 blocked └── receipt_batch_2026-03-12 ├── sc_9e8d7c6b5a4f... action: deploy_config A=1 ✓ └── ... Every entry is immutable. Every root is verifiable.
Certification

The SafeClash mark means one thing

Any system — a hospital's medication platform, a bank's transaction engine, a city's infrastructure controller — can apply for SafeClash certification. The process verifies that the system implements the full Governed AI OS protocol.

01 — Audit
Submit your system's governance implementation. We verify that the canonical loop is intact and that A=0 is mathematically enforced — not just claimed.
02 — Test
CLASHD27 runs the certification test suite — 1201 tests across 27 evaluation cells. Zero failures required. No exceptions.
03 — Certify
A SafeClash certificate is issued, anchored to the Merkle audit trail, valid for 12 months. Renewal requires re-audit.
Pricing

Pay for what you govern

Open
Self-hosted
Free — always
  • Full bootstrap included
  • Local receipt generation
  • Local Merkle audit trail
  • No certification mark
  • Community support
Enterprise
Certified
Contact — custom
  • Full certification audit
  • CLASHD27 test suite
  • Legal-grade receipts
  • SLA guarantee
  • Custom atom pricing
Ecosystem

Part of a governed whole

Foundation
governedaios.com
The philosophy and formula behind it all
Kernel
openclashd.com
Governance and consent decisions
Network
openclashgrid.com
Edge signals from the world
Certifier
safeclash.com
Trust, billing, the receipt