A managed customer-support employee

Hire your first AI employee.

Peaklight delivers.

  • Scoped in writing first
  • Human handoff
  • Audited console actions
peaklight console · support inboxIllustrative workflow
Incoming · Web chat · order #4187
"Where is my order? It was supposed to arrive Friday."
Policy check
action: order.status.lookup · source: Shipping policy v3, cited  Allowed
Agent reply · sent
"Your order shipped Thursday and is out for delivery today. Here's the tracking link. Sorry for the delay."
audit 14:02:11 · decision=allow · policy=support.v3 · record=evt_8c41
Incoming · Email · billing
"I was charged twice this month. I want a refund now."
Policy check
action: refund.issue · rule: refunds require human approval  Escalated
Handoff to human · with context
The conversation, evidence and a drafted reply are packaged for your team. The customer is told a person is taking over.
audit 14:05:37 · decision=escalate · reason=refund.human_approval · record=evt_8c42

What this is built on

Doctoral research
The sovereignty framework behind the product comes from ongoing doctoral research, which models an AI deployment across several distinct dimensions of capability and trust.
Governance in software
The rules that decide what an agent may send, what data may leave, and when a human must step in are enforced by conventional code, not by instructions given to a model.
A named operator
Every deployment has a person at Peaklight who runs it. You are hiring a managed service, not buying a self-serve tool and being left with it.

One request through Peaklight

  1. Customer askson a connected channel.
  2. Knowledge scopedto sources you approve.
  3. Rules decidewhat the agent may send and what needs your approval.
  4. Reply or handoffto your team, with context.
  5. Audit recordedfor every console and API action.
The role

An employee, not a chatbot.

You hire one role, scoped in writing, and a person at Peaklight runs it. It works inside limits you set and hands the difficult cases to your team.

Customer Support

Reports to
You
Hours
Always on
Channels
WhatsApp, plus two standard integrations
Managed by
Peaklight

What it owns

  • Answers customers on the channels they already use, in your voice, from sources you have approved.
  • Works inside limits you set: what it may say, what data may leave, and when it must stop.
  • Hands complex cases to your staff with the conversation and context already attached.
  • Records every message and console action so the work can be reconstructed afterwards.
  • Takes on more of the role only when you approve the expansion.
Sovereignty and governance

Governed in software, not by promises.

A model told to behave is not a control. Peaklight puts the decisions in ordinary code that runs before the agent acts, and writes down what it decided.

Decisions run before the action
Whether an action is allowed is settled by policy code, not by the model deciding to comply. A blocked action does not reach the customer.
Escalation is a rule, not a hope
Actions you mark as human-approval-only are handed to your team with the conversation attached. The customer is told a person is taking over.
Isolated per business
Each deployment runs in its own container with its own data, on infrastructure Peaklight operates rather than resells.
Providers are swappable in the architecture
The runtime talks to model providers through one interface, so a provider can be changed without rebuilding the deployment.
Actions are recorded
Console and API actions are written to an audit record, so what happened can be reconstructed after the fact.
Policy, illustrative
action: refund.issue
  requires: human_approval
  reason:   refunds are never agent-authorised

action: order.status.lookup
  allow:   true
  sources: [shipping_policy_v3]
  record:  audit_event

Read the research behind this: What sovereign AI actually means for a business

Onboarding

Scope, deploy, govern, grow.

  1. Scope

    We agree in writing what the role covers: the tasks, the channels, the voice, the sources it may draw on, and the lines it must not cross.

  2. Deploy

    We build and deploy the employee into its own isolated workspace, wired to your channels and the integrations in scope.

  3. Govern

    The limits go into policy code, not into a prompt. You see what was decided and why, and you set what requires a human.

  4. Grow

    Your operator tunes the role as real conversations come in, and widens it only when you approve the expansion.

Pricing

Simple monthly pricing.

One all-in price: the AI employee and the human operator who runs it. No add-on stacking, no per-conversation fees.

$700/mo
AI Employee

One managed AI employee, all-in: WhatsApp, two standard integrations, a human operator, safe mode, and a weekly report.

Hire your AI employee (AI Employee)
+$500/mo
Business Review

Adds an operator-produced monthly business review: attribution, sales sources, customer insights, and a monthly review call.

Hire your AI employee (Business Review)
Custom
Enterprise

The widest supported scope, defined case by case with your team.

Talk to Peaklight (Enterprise)

One-time setup fee of $1,500 covers your personalized build and integrations. Included scope is agreed in writing before you pay. Voucher credit sits on your account and applies to anything you owe.

Questions

What people ask first.

What exactly am I hiring?

One AI employee in one role, run for you by a person at Peaklight. The monthly price covers the agent, the runtime it runs on, and the operator who maintains it. It is a managed service rather than software you install and administer.

How do I stop it saying the wrong thing?

You approve the sources it answers from and the actions it may take on its own. Anything outside that is refused or escalated to your team by the policy layer, which runs before the reply is sent rather than relying on the model to hold back.

What happens when it cannot handle something?

It hands the conversation to your team with the history, the relevant evidence and a drafted reply attached, and tells the customer that a person is taking over.

Where does my data live?

Each deployment runs in its own isolated container on infrastructure Peaklight operates. What leaves that boundary is limited to the model calls the role needs, and the integrations in scope are the ones you agreed in writing.

Can I change model providers later?

The runtime reaches providers through a single interface, so a provider change does not mean rebuilding the deployment. It is a change we make for you as part of running the service.

What does it cost, and what is the setup fee for?

The AI Employee is $700 a month, all-in. The monthly business review is a $500 add-on. There is a one-time $1,500 setup fee for the personalized build and integrations, and the scope it covers is agreed in writing before you pay.

How do I start?

Send us a message. The next step is a scoping conversation where we work out what the role should cover and tell you honestly whether it is a fit. There is no self-serve signup: every deployment is built for the business it serves.

Get started

Reserve onboarding for your
first AI employee.

One role, governed from day one.

next step: a scoping conversation