Skip to content
LAUNCH FILM — LIVEGREENLIGHT · PATENT-PENDINGRUNTIME AUTHORIZATION FOR AI AGENTSMCP CONNECTORS — IN DESIGNISO/IEC 27001 — ROADMAPSOC 2 TYPE II — ROADMAPISO/IEC 42001 — ROADMAPMAS FEAT — DESIGN-ALIGNEDDETERMINISTIC · EXPLAINABLE · TAMPER-EVIDENTAI ACTS · CRELIS DECIDESLAUNCH FILM — LIVEGREENLIGHT · PATENT-PENDINGRUNTIME AUTHORIZATION FOR AI AGENTSMCP CONNECTORS — IN DESIGNISO/IEC 27001 — ROADMAPSOC 2 TYPE II — ROADMAPISO/IEC 42001 — ROADMAPMAS FEAT — DESIGN-ALIGNEDDETERMINISTIC · EXPLAINABLE · TAMPER-EVIDENTAI ACTS · CRELIS DECIDES
Crelis.ai

How Crelis works

The independent authority layer between AI and the real world

Before an AI agent takes a consequential action, Crelis evaluates it against policy and decides — allow, require human approval, escalate, or block — then issues or withholds the execution visa and records the evidence.

Request a Demo
A Crelis checkpoint: AI agents carrying consequential actions reach a gate that blocks a high-risk action, holds one for human approval, or allows it — and the allowed action receives a signed execution visa before it proceeds.
From intent to trusted action

Evaluate each action

Task type, proposed action, amount, industry and channel from the endpoint registration, customer tier, verifications and message signals are evaluated deterministically — before anything executes.

Decide by policy

Policy decides the outcome: allow, require human approval, escalate, or block. Thresholds are yours to set; the final decision is never made by an AI model.

Issue the visa

Allowed actions receive a short-lived, cryptographically signed runtime execution visa. Your system executes; the visa and the evidence make the approval and the execution the same record.

Seal the evidence

Each decision — inputs, policy version, reviewer sign-off, visa — is hashed into a tamper-evident audit trail you can export.

Decision engine

See the decision happen

Pick an action. Watch the inputs evaluated and the policy that fires resolve into a decision — allow, require human approval, escalate, or block. Illustrative data.

Decision engine · deterministic policy evaluation

Illustrative data

Incoming agent action — select one

Inputs evaluated

Action
payment.wire
Amount
$250,000
Industry
Financial services
Channel
Treasury API
Customer tier
Corporate
Verifications
Counterparty not yet verified
Policy appliedWIRE-HIGHWires above threshold or unverified counterparty

Require human approval

Visa withheld until an authorized reviewer signs off.

Policy WIRE-HIGH fired: amount above the wire threshold and counterparty unverified. Human approval required from a treasury reviewer; visa withheld until sign-off.

Recorded risk context91· context only, not the decider

Human escalation

A checkpoint, not a bottleneck

Require human approval is a first-class outcome. Authorized reviewers see the full context, and their sign-off becomes part of the execution record.

Human approval path

Illustrative data

  1. 01

    AI action

    Agent proposes a consequential action

  2. 02

    Deterministic policy

    Inputs evaluated; the policy that applies fires

  3. 03

    Require human approval

    Outcome when policy calls for sign-off

  4. 04

    Authorized reviewer decides

    Full context; sign-off joins the record

  5. 05

    Visa issued or withheld

    Signed execution visa; your system executes

  6. 06

    Evidence recorded

    Tamper-evident record written

Audit engine

Each consequential action leaves evidence

Who proposed it, which inputs were evaluated, which policy fired, who approved it, what executed — every decision reproducible from its recorded context and exportable for auditors and regulators.

Audit event stream · illustrative

Illustrative datasealed

  1. 14:02:11Z · Finance Agent

    Proposed wire transfer $250,000 → ACME-7741

  2. 14:02:11Z · Crelis

    Inputs evaluated · policy WIRE-HIGH fired · REQUIRE HUMAN APPROVAL · risk context 91

  3. 14:02:12Z · Crelis

    Routed to treasury reviewer pool (2 eligible) · visa withheld

  4. 14:06:48Z · Treasury reviewer

    Checked counterparty, approved with note

  5. 14:06:49Z · Crelis

    Execution visa issued · Ed25519-signed, single-use

  6. 14:06:49Z · Banking endpoint

    Visa verified · wire released — ref TXN-88231

  7. 14:06:49Z · Crelis

    Evidence recorded · hash 0x91f3…aa07

How this maps to Singapore's MAS SAFR white paper: our breakdown of the records SAFR implies

Governance

One pane for AI activity

Approvals, exceptions, and compliance events across the agents you route through Crelis.

Governance · todayIllustrative data

all policies active

48,212

AI actions today

94.6%

Allowed

2,431

Human approval required

17

Blocked

Open exceptions

  • Ops Agent · bulk delete · blocked
  • Finance Agent · wire · human approval required
  • Compliance Agent · KYC override · escalated

Platform roadmap

Built as an operating system for trusted execution

The trust layer is the foundation. These modules are planned capabilities that extend the core authorization layer.

MCP integrations

In design

Model Context Protocol as an integration surface — connectors in design so agent frameworks can route consequential actions through Crelis.

AI agent registry

In design

Identity, permissions, and decision history for each agent you register with Crelis.

Human marketplace

In design

Optional, future extension: a pool of authorized reviewers an organisation could draw on when policy requires human approval.

Enterprise dashboards

Pilot

Org-wide visibility into AI activity, approvals, exceptions, and spend.

Compliance controls

In design

Policy packs built on over 500 native control templates; mappings to regulatory frameworks are in design.

Audit engine

Pilot

Tamper-evident execution records with cryptographic sealing and export.

Workforce orchestration

In design

Routing, SLAs, and quality scoring across authorized reviewer pools.

Apply for Pilot Access