The GREENLIGHT platform
One platform for authorizing what AI agents do.
GREENLIGHT defines what agents may do, decides on every consequential action, enforces the decision and proves what happened — connected to the agents you already run. One product, five capability areas.
How GREENLIGHT worksPolicy
Define what agents may do
Native rules and your own, versioned and protected.
Decide
Evaluate every consequential action
Intent and risk scanned; one of four outcomes, with reasons.
Enforce
Act on the decision — and only on the decision
Human approvals and execution visas, single-use and action-bound.
Evidence
Prove what happened, to anyone who asks
A tamper-evident logbook, verifiable evidence and reporting.
One decision, end to end
What happens when an agent wants to act
Every consequential action an agent proposes passes through the same four steps before it reaches your systems.
01
Agent proposes an action
Before a tool call reaches your system, it goes to GREENLIGHT through the gateway, a hook or the SDK.
02
GREENLIGHT decides
Your policies and the native library are applied to the action, its target, the amount and the context.
03
The decision is enforced
Allowed actions carry an execution visa. Others are held for a human, escalated or blocked.
04
Evidence is kept
Every decision, approval and visa is written to a signed logbook you can verify independently.
Four outcomes, always with a reason
Allow
Inside policy. The action proceeds with an execution visa bound to what was approved.
Require human approval
Consequential. A named approver decides, with separation of duties and a time limit.
Escalate
Unfamiliar. The rules do not recognise the action, so it goes to a person instead of passing silently.
Block
Against policy. The action does not reach your system, and the reason is recorded.
Capability areas
Everything GREENLIGHT does, by area
Each area is part of GREENLIGHT — nothing is sold separately.
Native policy library
A library of 550+ native rules for consequential actions — payments, refunds, account changes, data movement — active from day one.
Policy Studio
Write organisation rules in plain language, see what they would change before they go live, and promote them from UAT to production with approval.
Protected floor
Native protections can be tightened, never weakened. Where a threshold is designed to be tuned, you tune it — the floor itself stays in place.
Intent & risk scanning
Each proposed action is described consistently — what it does, to which system, for how much and on whose behalf — and scanned for risk signals before any rule runs.
Decision engine
Every action receives one of four outcomes: allow, require human approval, escalate or block — with the reasons, the deciding policy and how well the action was covered.
Read safety
Reads are observed rather than blocked by default, and screened for bulk extraction, sensitive data and outbound movement.
Rollout modes
Start in shadow — Crelis records what it would decide and blocks nothing — then move to advisory and enforce when the evidence supports it.
Human approvals
Held actions go to a reviewer with the context to decide — in the Crelis console or inside the tools you already use, such as ServiceNow and Jira.
Execution visas
Every allowed action carries a signed, single-use, short-lived visa bound to that exact action. The receiving system checks it before it writes.
Exceptions
When a business needs a temporary exception, it is requested, approved and time-boxed — never a silent bypass.
Audit logbook
A tamper-evident record of every decision, approval and release. Any later edit breaks the record — and shows.
Verifiable evidence
Signed checkpoints let an auditor verify the record offline, without having to trust Crelis, and evidence can be exported to write-once storage.
Analytics & reporting
Volume, outcomes and trends over 14, 30 or 90 days, coverage of what your agents actually do, escalation SLAs and redacted reports.
SIEM export
Stream decisions and approvals into your existing detection and response stack.
Customer-hosted gateway
A gateway you run in your own environment that puts agent tool calls through GREENLIGHT before they reach your systems. It only opens outbound connections.
Agent & coding-agent hooks
Pre-action checks for seven agent frameworks and for coding agents such as Claude Code and Cursor.
SDKs & decision API
Python and TypeScript SDKs and a REST decision API for agents and applications you build yourself.
Connector library
A library of MCP servers and AI providers, each reviewed for decision-safety before it is listed.
Where it fits
The layer between an agent’s intent and your systems
GREENLIGHT does not replace the controls you already run. It adds the one they leave open: a decision on each action, before it happens.
Identity & access
Decides who an agent is and which systems it can reach.
Model guardrails
Filter what a model reads and writes: prompts, outputs and content.
GREENLIGHT
Decides whether a specific action should happen now, holds it for a person when needed, and proves the decision.
Monitoring & SIEM
Watches what happened and alerts on it. GREENLIGHT feeds it every decision.
Enterprise foundations
Built for regulated environments
Hosted in Singapore
The decision service runs in Singapore.
Sign-in your way
Sign in with Microsoft and passkeys for administrators.
Role-based access
Partner → customer → tenant roles with tenant isolation.
Separate environments
UAT and production kept apart, with promotion between them.
Scoped keys
API keys limited by scope and expiry.
Your credentials stay yours
Nothing calls into your network.
Get started
Start with one workflow.
Pick the agent workflow that worries you most. We will connect it in shadow mode, show you what GREENLIGHT would decide, and agree with you when to enforce.