NobleCloak
Approvals for AI agents · open-source plugins

Add the plugin to your agent. Policy decides which actions need a person. The request reaches the right approver in Slack or the web. The agent waits and resumes exactly once, and you keep the Record of Decisions.

  1. proposed · wire_transfer $48,200
  2. awaiting · m.alvarez (Finance) · Slack
  3. approved · resumed once · recorded
// one line per risky action
wire_transfer   $48,200   needs approval
send_email      external  needs approval
read_ledger     ar-close  allowed

// policy picks the approver, not your code
// the run pauses, a person answers, it resumes once
Plugins

Open source, in the runtime you already use.

Apache-2.0 plugins and adapters. Your runtime keeps its own pause; the plugin asks us who decides and hands back the answer. One line per risky action.

  • LangGraph

    TypeScript · Python coming soon

    Available
  • Claude Agent SDKat launch
  • Claude CodeAvailable
  • Any MCP client

    Through the gateway

    Available
  • Vercel AI SDKComing soon
  • OpenAI Agents SDKComing soon
  • Strands / AgentCoreComing soon
  • Google ADKComing soon
How it works

One approval, start to finish

  1. 1

    The agent proposes

    Your agent reaches an action that touches money, customers or production. Policy, not your code, decides whether it needs a person, and which person.

    ar-close · run 7f3a · 13:59:02
    action
    wire_transfer
    to
    Harbor Supply
    amount
    $48,200.00
    policy
    Wire transfers need Finance approval
    state
    Run paused on its own checkpoint
  2. 2

    A person decides

    The right approver gets the request where they already work: Slack or the web inbox, with Teams and email coming soon. An approval covers one action, never the rest of the session, and a sub-agent never inherits its parent's approval.

    Slack · #finance-approvals
    NobleCloak needs a Finance approver for one action from ar-close: wire $48,200.00 to Harbor Supply.
    requested by
    D. Park (requester, cannot approve from Slack)
    choices
    Approve this action · Deny
    scope
    Covers this action only
  3. 3

    The record stays

    After a named approver decides, the agent itself resumes, exactly once. Prompt, action, policy, approver and outcome become one line in the Record of Decisions. Searchable on every plan; export comes with Business and Enterprise.

    Record of Decisions · rec_01HZ4Q7M…
    • Prompt: Close out AR for September…
    • Action: wire_transfer · $48,200.00
    • Policy: Wire transfers need Finance approval
    • Decision: approved · M. Alvarez · 14:02:11
    • Outcome: Resumed once · wire sent · gateway observed
Your runtime already gives you

The pause is built in. The people are not.

A durable pause. An allow or deny check. A boolean hook per tool.

What it leaves to you, until now
  • —Who may approve, and who is asked
  • —Reaching that person where they already work
  • —Applying the answer to the action that is waiting
  • —One policy and one record across every surface
Why we built this

Who may act, and who must ask.

Every business operates on that simple fact, but AI agents don't have a way to learn the company structure and rules other employees do in onboarding. We are building the tools to let agents work across your organization with the same approvals and records as their human teammates.

The Record of Decisions

Every prompt, the action it caused, and who said yes.

Searchable on every plan; export and SIEM feeds come with Business and Enterprise. Entries say whether the gateway saw the call or the runtime reported it. Longer history comes with the plan; examiner evidence packs are coming.

TimeActorActionDecisionByAttestation
14:02:11ar-closewire_transfer $48,200approvedm.alvarezobserved
13:47:02support-botrefund $1,150deniedj.okaforobserved
13:40:58researchsend_email (external)awaitingFinance roleasserted
13:12:30ar-closeread_ledgerallowedpolicyobserved
Deploy

One API. Designed to run where your data must live.

  • Hosted

    Our control plane, your runtime.

    Available
  • Dedicated tenant

    Your own database, private network option.

    Beta
  • Edge, in your cloud

    The same service, run from your own cloud account.

    Coming soon
Pricing

Pick a plan. Users are included.

Annual billing: two months free on Team and Business.

Free

$0
  • One builder gating a first action
  • Plugins, adapters, web inbox
  • 7-day history
Get started

Team

$150/mo
  • 10 users included
  • The agent gets shared across the team
  • Slack delivery (Teams coming soon)
  • 30-day history
Get started
Most teams

Business

$500/mo
  • 50 users included
  • Role-based policies
  • SSO (coming soon)
  • 1-year history
  • Audit export
Get started

Enterprise

By organization size

$24k to $90k/yr
  • Gateway policy across Claude, org-wide
  • ChatGPT (coming soon)
  • Starts with the AI Trust Report
  • Up to 7-year history
  • AI risk tracking
  • Proof and examiner evidence packs (coming)
Talk to the founders

A user is anyone with a login. Need more users? Move up a plan.

Regulated

Examined? Start with the AI Trust Report.

A read-only assessment of the AI tools your own organization uses, packaged for an examiner. Built for banks, credit unions and RIAs to get a first answer, then grow into Approvals on the Enterprise plan.