The authorization layer for AI agents

Your AI agent just took an action. Who authorized it?We make sure someone always does — and prove it, forever.

The stakes

Every consequential action becomes a legal question.

INCIDENT · 03.14

33,840 emails. One approval missing.

A support agent's autoreply loop reached the entire customer base before anyone at the company saw it.

INCIDENT · 06.02

$1.7M returned to the wrong accounts in eleven minutes.

$2.3M returned to the wrong accounts in eleven minutes. No human ever saw the parameters.

INCIDENT · 09.19

4 weeks of downstream reconciliation.

Six weeks of downstream reconciliation. No record of who — or what — authorized the change.

How it works

Three moments. One unbroken record.

STEP 01

Pause.

The agent requests a consequential action. Ceadly intercepts before execution.

@ceadly.guard(policy="prod_write")
agent.execute(action)

STEP 02

Review.

The decision routes to the right human under your policy. They see everything. They approve, deny, or amend.

→ reviewer: risk-officer
→ sla: 4m 12s
→ decision: approved

STEP 03

Resume.

The action executes. The record is hashed, signed, and committed to a public ledger. It can't be altered later.

sha3-256: 4c9d...ae12
ledger: opentimestamps.org
height: 872,145

Proof of scale

Every decision that touches Billing.

Core capability

Six primitives. One governed system.

01

Human-in-the-loop.

Every consequential action pauses. A named person decides. Nothing bypasses that gate.

02

Policy routing.

Rules decide who reviews what. By dollar amount, by system, by risk tier. Enforced identically at 3am and at scale.

03

Tamper-proof records.

SHA3-256 hashes. Daily commitments to a public ledger. What happened can be proved. What didn't, can be disproved.

04

Oversight dashboard.

Every pending decision, every past record, every reviewer's action — one interface, one source of truth.

05

Governance evidence package.

Auditor-ready exports. SOC 2, HIPAA, EU AI Act, sector regulators. Generated on demand.

06

One decorator.

@ceadly.guard() above any function. That is the integration.

Pricing

Priced by consequence, not seats.

DEVELOPER

Free

1,000 events/month — enough to run one production agent end-to-end before you ever talk to sales.

  • Self-serve
  • One reviewer seat
  • 90-day record retention
  • Community support
GROWTHRECOMMENDED

$0.002/event

Unlimited events. You pay for what your agents actually do, not for a seat count that has nothing to do with risk.

  • Unlimited events
  • Policy engine
  • SSO
  • SLA
ENTERPRISE

Custom ACV

Flat monthly, unlimited events, SSO, RBAC, SOC 2, full Governance Evidence Package. Priced to agent fleet size and regulatory exposure — a quote within 48 hours.

  • Everything in Growth
  • Private ledger commitments
  • Dedicated auditor liaison
  • On-prem and air-gapped options
ADD-ON · IMMUTABLE EVIDENCE

$2K–$10K/mo

The range tracks event volume requiring cryptographic sealing. $2K covers a single regulated workflow. $10K covers your entire agent fleet, sealed and litigation-ready.

  • Cryptographic sealing add-on
  • SHA3-256 daily commitments
  • Public-ledger anchoring
  • Litigation-ready evidence exports

Take control

Your agents will act. Decide who lets them.