HUMANLATCH by VerdictLayer

Approval control plane for AI agents

Start with cloud, not a six-month rollout

Stop explaining the product. Let people protect one real action in the next 10 minutes.

HumanLatch sits between agent intent and execution. Start in the hosted cloud product, define one approval rule, and route the first risky action through a real human checkpoint without building a governance platform from scratch.

Hosted cloud first, self-host later if procurement requires it

One policy, one API key, one test action to prove value

A mobile-friendly approvals experience when nobody is at a desk

Fastest path to value

Protect one consequential cloud action first

Step 1

Create a cloud workspace

Start in the hosted product without setting up infrastructure or internal tooling first.

Step 2

Define one approval rule

Pick a single sensitive action like prod deploys, refunds, or factory overrides and require review.

Step 3

Send a test action

Watch HumanLatch auto-approve, escalate, or block the action immediately in the dashboard.

What the first win looks like

terraform.apply to prod

The agent prepares the change, but HumanLatch decides whether it runs now, waits for review, or stops.

RouteNeeds approval

Production context pushes the request into approval. A teammate can review it from the hosted dashboard and keep the trace for later.

Why this is easier

The onboarding path is one narrow job, not a giant platform decision.

Teams do not need to model every workflow on day one. Connect the one action that makes everyone nervous, define the route, and prove the checkpoint works in production-like conditions.

Step 1

No infrastructure setup is required for the hosted cloud start.

Step 2

The dashboard shows the route clearly: approved, pending approval, or blocked.

Step 3

Approvals and traces stay visible for operators and stakeholders.

Step 4

Self-hosting can come later when legal or security requires it.

Start here

Pick the first action where a mistake would cost real money, uptime, or safety.

HumanLatch is strongest when an AI system is no longer just generating text and is about to change a real system.

Cloud

Gate deploys, IAM, and infrastructure changes.

Put human approval in front of the few actions that can actually break production.

Finance

Review refunds, credits, and account changes.

Let agents move fast on low-risk tasks without handing them open financial authority.

Operations

Protect robotics, factory, and fulfillment overrides.

Keep a visible decision point before an AI action touches physical systems or irreversible workflows.

Start now

The default path should be obvious: use the hosted cloud product first.

Create a workspace, add one policy, and send one action through the approval flow. If the team needs a deeper security review later, move to self-hosting after the value is already clear.