VERDICTLAYER

Oversight infrastructure for AI

Human control for consequential AI

Put a decision layerbetween AI and execution.

VerdictLayer builds the control surface that lets teams ship agentic systems without surrendering operational judgment. Agents can move quickly. High-stakes actions still need a verdict.

Decision outcomes

approve, escalate, block

Best first buyer

infra, ops, security

Go-to-market

free CE plus $99 cloud

HumanLatch

Approval control plane for AI agents

Shipping now

What it does

Intercept risky agent actions before they execute.

HumanLatch accepts a proposed action, evaluates policy and risk, routes it for auto-approval, review, or blocking, and writes the entire outcome to an audit trail.

Operator clarity

Decision traces explain why an action was allowed, escalated, or stopped.

Commercial fit

Strongest initial pull is from infra teams already worried about unchecked automation.

Example

`terraform apply` against production cluster

Verdict: `pending_approval` because of production scope and destructive potential.

Thesis

The model is not the whole system.

Once AI can trigger real operations, the missing layer is not another benchmark or another wrapper. It is the decision boundary between intent and action.

01

Intent before execution

Agents propose what they want to do before they do it, so policy can evaluate the action at the right moment.

02

Policy with consequences

Rules should not be advisory. They should decide whether the action can run, needs a person, or must fail closed.

03

Fast human review

Only the actions that matter should reach operators, with enough context to make a defensible call quickly.

04

Audit you can defend

Every verdict, reviewer action, and downstream outcome becomes part of the record instead of disappearing into logs.

Products

Start with the sharpest commercial wedge.

HumanLatch is first because it solves the most immediate buyer pain: preventing dangerous AI actions from executing without oversight.

HumanLatch

Available now

Approval control plane for AI agents

One proposal API

Wrap high-stakes actions once instead of inventing approval logic inside every workflow.

Decision trace

Show the exact policy match, risk factors, and audit timeline when a buyer asks why a verdict happened.

Hosted or self-hosted

Adopt CE in buyer-controlled environments or move to cloud when speed matters more than owning the stack.

Best first buyers

Infra, ops, and security teams where one bad autonomous action creates immediate pain.

Signals

Next

Behavioral anomaly and drift signals across fleets of agents, tools, and operators.

Policies

Roadmap

A broader enforcement layer spanning multiple products, domains, and execution surfaces.

Audit

Roadmap

Forensic timelines and export surfaces for teams that need compliance-grade visibility.

Use cases

Where oversight becomes a budget line.

Infrastructure agents

Terraform applies, IAM changes, credential operations, and deploy-time actions.

Operational automation

Refund thresholds, account changes, and workflows with customer or financial impact.

Internal copilots

Assistants with tooling access that can trigger real changes instead of just generating text.

Physical systems

Robotics and manufacturing flows where a checkpoint is mandatory, not optional.

Pricing

Free to adopt. Obvious path to paid.

Community

Free

Self-hosted forever

The full HumanLatch core in your own environment. MIT licensed and buyer-controlled.

  • Approval control plane
  • Policy engine
  • Audit trail
  • Single workspace
  • Docker deployment
Self-host on GitHub
Fastest path to revenue

Cloud

$99

per month

Managed hosting for teams that want operational oversight without running the control plane themselves.

  • Up to 10 team members
  • 10,000 monthly actions
  • 90-day retention
  • Email notifications
  • Outbound webhooks
Start free trial

Enterprise

$499

per month

For larger teams that need deeper governance, longer retention, and rollout support.

  • Unlimited team members
  • 1M monthly actions
  • 365-day retention
  • Slack approvals
  • Advanced audit exports
Talk to us

Give AI systems room to operate.Keep the final say where it belongs.

If your agents can deploy, refund, approve, modify, or override, the oversight layer is part of the product, not an optional add-on.

Talk to VerdictLayer

Enterprise pricing, buyer questions, partnerships, or rollout support.