- Ask about any client you work on“How did Acme’s leads do since the promo ended?” From your own Claude or ChatGPT, it runs your playbook on their accounts, your call notes and your benchmarks.
- Only the clients you’re onEach person on your team signs in as themselves and reaches only the clients they’re assigned. Owners and admins see every client.
- Your team’s answers stay with your teamThe client never sees them, and they don’t count against the client’s own questions.
For agencies
Your agency’s know‑how, in every AI your team and clients use.
Your team asks about any client from their own Claude or ChatGPT. Each client asks about their own business, in your name. Connect the logins you already have. Nothing to build or host.
No codeEach client sees only their ownFrom $99 a month
Northstar AgencyResults Review
v1.3Example client · Acme
Client asks
Sales are down this month.
Is the marketing working?
Not a marketing problem
Fewer visitors, not worse ones.
People who came bought as usual. Visits fell from one day, the 9th, so find what changed that day before changing anything else.
What it checked
- Visits
- 17% lower since the 9th
- Share who bought
- Same as usual
One setup, two sides
Your team on every client.
Each client on their own.
- Your answer in their own AIThey add it to their own ChatGPT or Claude in about two minutes and ask whenever they like. They get your team’s answer, in your name.
- A weekly review in their inboxFrom your agency, on their numbers. Your team approves each one until you trust it to go out on its own.
- Only their ownTheir accounts and the notes you mark for them. Never another client’s data, never your internal notes.
What your client sees
Your client asks a question.
They get your team’s answer.
What the numbers show, what it means, and what to do next, the way your team would put it. And what it can’t tell yet.
Results Review
Not a marketing problem
What the numbers show
- Sales are 15% lower than the same days last month.
- Visits to the site are 17% lower, and the drop starts on the 9th.
- The share of visitors who buy is the same as usual.
What this means
The site still sells as well as before. Fewer people are reaching it, and a drop that starts on one day usually means something changed that day, not that the marketing stopped working.
Limit: These numbers show where sales dropped, not why. Someone has to look at what changed on the site that week.
Based on: website analytics and sales for this month and last
Dates: September 1–20, 2026, against August 1–20
Dates: September 1–20, 2026, against August 1–20
What it knows
Everything your agency knows about each client.
In one place.
The accounts it reads
Accounts are only read: nothing changes a campaign, a budget or a record. Store and CRM numbers arrive as totals: orders, sales, new contacts and deals by stage. Never a customer’s name, email or phone. Each client’s review reads only that client’s accounts, never another client’s. Pages for your kind of agency: HubSpot, Shopify, SEO, Local, Klaviyo.
Who sees what
Built for client data.
Not patched on later.
The questions to ask before any AI touches a client’s accounts, answered before you sign up.
Everyone signs in as themselvesYour team and each client, with their own login. A team member reaches only the clients they’re assigned.
Each client, only their ownA client’s AI reads that client’s accounts and the documents you marked for them. Never another client’s.
Internal notes stay internalEvery document starts as your team’s alone. You choose what a client’s AI may read.
Read only, and nothing goes out on its ownAccounts are only read. Your own AI can’t change anything until you turn changes on, and whatever it would send a client (a reply, an invite, a new version) waits for you to approve it.
Every call is keptWhat each person’s AI asked, what each client asked, and which documents each answer read.
Remove someone, access endsSomeone who leaves your team, or a client you remove, loses access at once.
Your work, seen every week
Clients don’t want another report.
They want your answer at 9am Monday.
Work a client can’t see gets cut at renewal. A review that lands in their inbox every week gets seen.
1 · SchedulePick the review and the day
Weekly, every two weeks or monthly, on the day and hour you choose.
2 · RunIt runs on their numbers
From the accounts you matched to that client and the notes you keep. The client doesn’t sign in to anything to get it.
3 · ApproveYour team signs off
Read it, add a note, send or skip. Approve every one, only the first, or none once you trust it.
4 · LandIt arrives in your name
From your agency, and replies come to you. A review nobody approved in three days isn’t sent.
Every review adds up on the client’s own page.
What you sent, what they asked, and a list of things to fix that either side can mark done. It opens from a private link you drop into an email or your monthly deck, on your own domain. No account needed.
Acme · from Northstar Agency · Example
Things to fix
- Find what changed on the site on the 9thWe’re on it
- Contact form doesn’t send on mobileOpen
- Put back the tracking tag the redesign removedFixed
Reviews sent
- Results Review · Monday, Sept 21Sent
- Results Review · Monday, Sept 14Sent
The question behind the other questions
“Won’t this teach clients they don’t need us?”
They’re asking ChatGPT already. Today it answers with whatever it finds and tells them they’re right. This makes it answer the way your team would, in your name.
Your name on every answerYour agency’s review, your playbook version, your weekly email.
Your team still makes the callIt only reads accounts. Changing anything is still your team’s job.
It says when it needs youWrite it to name what’s missing instead of guessing, and to send the client back to you.
At their monthly limit, back to youPast their monthly questions, their AI tells them to check with you.
Build it, or switch it on
Hosting it is cheap.
Building it and keeping it safe isn’t.
A server your team’s AI can call costs a few dollars a month to run. The work is everything around it: signing each person in, keeping every client’s data apart, keeping a record of what was read, and keeping it working as ChatGPT and Claude change. That part is ours.
| Option | What you build and keep running | Can the client copy your playbook? | When a client leaves |
|---|---|---|---|
| Build your own server | Sign-in, keeping clients apart, records, every connection, and fixes when AI apps change | Depends how you build it | Depends how you build it |
| Sharing a custom GPT | Nothing, until custom GPTs stop running on December 11, 2026 | Often: instructions and files can be coaxed out | They keep it until you stop sharing it |
| Sharing a Claude Project | Nothing, but only members of your own Claude organization can use it | Members can read it | Access ends once you remove them |
| ExpertRelay | Nothing: connect your logins and your notes | No: they never receive the files | Access ends the moment you remove them |
For the technical person on your team: each client gets their own hosted MCP server, your team gets one of its own, both with OAuth sign-in, plus a REST API and webhooks for your other systems.
Where ExpertRelay doesn’t fit: clients on free ChatGPT can’t add it yet (they need Plus or Pro). It doesn’t read databases, Google Sheets or shared drives yet. It doesn’t build website chat widgets, AI receptionists or automations that change campaigns; it only reads a client’s accounts. And useful answers can still reveal parts of how your agency thinks.
Start small
Switch it on for one client
before anyone else sees it.
Connect what you have
Sign in with your manager logins, add one client without inviting them, and drop in a few of their call notes.
Nothing to build.Ask it yourself
From your own Claude or ChatGPT, run your playbook on that client, including the messy cases with missing or conflicting numbers.
No client sees it yet.Turn it on for them
Invite them, or start the weekly review. If it earns its place, add the next client.
You set the price.Pricing
$99 a month for your agency.
$49 for each client you turn on.
No setup fee. Free for 14 days. Your team works on your clients on the agency plan; turn a client on when they should get their own AI. We bill you $49 per client you turn on. You set what they pay: bill it as its own line, make it the reason for a higher tier, or keep it in the retainer.
Agency plan
Start here$99 / month
Your agency’s know-how in your team’s own AI, on up to 25 clients.
- Your whole team, each on the clients they work with
- Your own AI connection and API
- Each client’s accounts and documents, and your library
- Your playbooks run on any client, 100 times per client a month
- Weekly reviews for your team
14 days free. A card starts the trial; nothing is charged until it ends.
Each client you turn on
+$49 / client / month
When a client should get your answers themselves. Turn one on or off any time.
- Their own AI, in their ChatGPT, Claude or Cursor
- Weekly reviews in their inbox, in your name
- Their own page: reviews, questions, things to fix
- 300 of their own questions a month
- Never another client’s data, never your internal notes
Turning one off lowers your bill; your team keeps everything.
Clients use Claude, ChatGPT (Plus or Pro) or Cursor.
More than 25 clients?
Talk to us about volume, support, and which tools you need connected.
Practical questions
Before you switch it on for a client.
Can my team use it on every client, from their own AI?
Yes. Each person connects their own Claude or ChatGPT once and works on the clients they're assigned; owners and admins see every client. The agency plan covers your team's work on up to 25 clients. Answers your team asks for go to your team only and don't count against the client's own questions. Each client allows your team 100 a month.
Can my clients use it inside their own ChatGPT or Claude?
Yes. Clients connect from their own Claude, ChatGPT or Cursor account in about two minutes, and your reviews answer right there in the chat. ChatGPT clients need a Plus or Pro plan; free ChatGPT accounts can't add it yet.
Do we have to build or host anything?
No. There's no server to run and no code to keep up. You sign in with the logins you already have, add your notes, and choose which clients get which playbooks. Keeping it working as ChatGPT and Claude change is our job.
What does it read for each client?
Their ad accounts (Google, Meta, LinkedIn, TikTok, Microsoft, Pinterest, Snapchat, Reddit, X, Apple Search Ads, Criteo, Taboola, Display & Video 360 and StackAdapt), their Google Analytics 4, Search Console, Google Business Profile and Klaviyo, and their Shopify sales and HubSpot contacts and deals as totals, never customer details. Plus the documents you keep about them (call transcripts, research, briefs) and your agency library. Your team signs in once with the access it already has and matches each account to its client. It only reads; it never changes a campaign, a budget or a record.
Can a client's AI see our internal notes?
No. Every document starts as your team's alone. A client's AI reads only the documents you mark for that client, and only the library documents you mark for clients.
What does a client get in their inbox?
The review your playbook gives for their numbers that week, sent in your agency's name, with your team's note if you added one. Replies come to you. You choose the review, the day and hour, and how often: weekly, every two weeks or monthly. Approve every one, approve only the first, or let it send on its own once you trust it; a review nobody approves within three days isn't sent.
Can clients copy our playbook or prompts?
No. Clients never receive the files behind a review; they get the answers. Useful explanations can still reveal parts of how your agency thinks, so choose deliberately what each review returns.
What happens when a client leaves?
Remove them and their AI loses your reviews at once: the connection stops answering, stops reading their accounts, and their seat comes off your bill. Your playbooks, your notes about them and the history of what they asked stay with you; they never received the files. The client keeps their own chats and the answers they got, which live in their AI app, not with us. Their invite page tells them this up front.
How is billing worked out?
$99 a month for the agency plan: your whole team, your own AI connection and API, your library, and your team's work on up to 25 clients. Then $49 a month for each client you turn on for themselves: their own AI, weekly reviews in their inbox and their own page. No setup fee. The first 14 days are free; a card starts the trial and nothing is charged until it ends. Turning a client on raises the bill by $49 a month, prorated; turning one off or removing them lowers it. Each client you turn on includes 300 of their own questions a month; past those, their AI asks them to check with you until the month resets.
How is this different from a Custom GPT, a Claude Project or a white-label chatbot?
A shared Custom GPT's instructions and files can often be coaxed out, sharing one isn't billing per client, and OpenAI retires custom GPTs on December 11, 2026. A Claude Project can only be shared with members of your own Claude organization. A white-label chatbot is a separate chat window under your brand. ExpertRelay works inside the ChatGPT or Claude your clients and your team already use, keeps each client's data apart, keeps the files on our side, and ends access when you remove someone.
What should we charge?
We don't know your clients, so we don't suggest a number. Price it as access to your team's thinking between calls: a monthly line per client, or the reason for a higher tier, and a one-time setup fee if you load their accounts and past call notes for them. Or keep it in the retainer to show the work between reports. Try it with one client before offering it to the rest.
Start with one client
Your know-how,
in their AI.
Connect the logins you already have, ask it about one client yourself, then decide. 14 days free.
Try it on one client, free