SOREN · THE AI MEDIA-BUYING AGENT FOR META ADS

An agent that answers with the fix, not a chart.

Ask anything about your Meta account and the conversation ends in an artifact: a diagnosis with the evidence cited, an action you approve, a report it builds, a run you schedule. Built for media buyers and the agencies behind them.

ONE REAL INSIGHT BEFORE YOU PAY · 30-DAY GUARANTEE
Workspace · greenfield-coffeeSAMPLE
ASK · A SCOPED QUESTION
◈ Prospecting-Broad

ROAS fell off a cliff this week but CTR looks fine. What’s going on?

Ran diagnosis · 4 passes · evidence attached

It’s the auction, not the creative. Reels took over the placement mix: CPM +28% while CTR held flat.

DIAGNOSIS · PROSPECTING-BROADRELIABILITY HIGH
WHAT

ROAS 3.2× → 2.1× (−34%) over 7 days. Spend flat. CPM $13.90 → $17.80 (+28%). CTR steady at 1.2%.

WHY

Auction pressure from a placement mix shift: Reels went from 18% to 41% of spend. CTR held, so it isn’t the creative.

SO WHAT

Every result costs ~28% more media while this runs, about $570/week at current pacing.

NOW WHAT

Pin placements to Feed + Stories for 48h: pause the Advantage+ placements ad set. Watch: CPM back under $15.

Pause Advantage+ placements ad set · Prospecting-BroadRECORDED ON APPROVE · UNDO READY
RUN · FROM THE PROMPT LIBRARY
Runs

Build the wrap for a flight — client-ready, exportable

Built wrap deck · 92-day flight · evidence appendix attached
Summer Cold Brew · campaign wrapDECK · PPTX / PDF
SCHEDULE · RUNS BECOME A WEEKLY REPORT CADENCE
⏱ Schedule this · Weekly cadence reportScheduled: Mondays 08:00 · builds itself every Monday
YOU APPROVE EVERY CHANGE
FOUR WAYS TO USE IT

Ask. Run. Schedule. Approve.

Each one is a way you already work: ask about one campaign or one week, run a pre-built job like an account review, put that job on a weekly schedule, approve what it recommends.

ASKScoped questions

Scope any campaign, ad set, or window and ask. Answers end in artifacts: a diagnosis with evidence, not a paragraph.

RUNThe prompt library

Pre-built asks for the recurring jobs: account review, deep dive, wrap deck, cadence report. Heavy ones run durably and return a deliverable.

SCHEDULERuns become cadences

Recurring asks (the account review and the weekly cadence report) can be put on a schedule and build themselves.

APPROVEActions on your yes

Pause / resume style status changes only, each gated on your explicit approval, recorded and reversible.

Get my first insightSee the live demo
ONE REAL INSIGHT BEFORE YOU PAY · 30-DAY GUARANTEE
WHAT SOREN WON’T DO

It diagnoses. You decide.

Reads by default

Sentrum connects via Meta OAuth and never changes anything by itself. It reads your performance data to diagnose. That’s it, until you say otherwise.

You approve every change

The only changes it can ever apply are ones you explicitly approve: pause or resume, a schedule change, or a new ad created as a PAUSED draft. Budgets, bids, and targeting are off-limits.

The log is the report

Every action is tracked, reversible, and attributable: an audit trail you can hand a client. No black box, and you can revoke access anytime.

Run Soren from Claude, Cursor, or any MCP client.

Sentrum ships a first-class MCP server: diagnose campaigns and ask follow-ups from the AI assistant you already use, no new dashboard to log into.

Connect to Claude
FREQUENTLY ASKED

Questions before you connect.

Who is Soren?

Soren is Sentrum’s diagnostic analyst, the voice behind every insight. Ask a plain-English question about your account and Soren answers from your real performance history: your baselines, your optimization log, your past winners. Every claim cites the data behind it; when the data is thin, Soren says so instead of guessing.

Can Soren change anything in my ad account?

Not on its own. Sentrum connects with read and management permissions, and the only changes Soren can apply are ones you explicitly approve: a pause or resume, a schedule change, or a new ad created as a PAUSED draft. Budgets, bids, and targeting are off-limits; there is a standing never-touch-budgets rule in the product itself. Every action is logged and reversible, and you can revoke access anytime.

What can I actually ask?

Anything about your Meta account, four ways: scope a campaign, ad set, or window and ask a question; pick a pre-built run from the prompt library (account review, deep dive, wrap deck, cadence report); put a recurring run on a schedule so it builds itself; or approve an action Soren proposes. Answers end in artifacts (a diagnosis with evidence, a deck, a scheduled report), not a paragraph.

Is this just a chatbot on top of my data?

No. A chat that only describes your account is a reporting tool with a text box. Soren operates on it: the conversation ends in a diagnosis with the driver named and the evidence cited, an action gated on your approval, or a client-ready deliverable. When the data is thin, it says so instead of guessing.

How long does setup take?

About two minutes, with the same Meta OAuth login you have done a hundred times. Sentrum starts pulling in your historical data right away, and the daily sweep takes it from there. No implementation cycle, no consulting hours.

Can I use Soren from Claude or Cursor?

Yes. Sentrum ships a first-class MCP server with diagnostic tools from getCampaignDiagnostics to planMedia. Diagnose campaigns and ask follow-ups from Claude Code, Claude Desktop, Cursor, or any MCP-compatible client, without opening a new dashboard.

GET STARTED

Put an agent on the account by tomorrow morning.

Connect your ad account in about two minutes. Soren starts on your real history the same day, and never changes a thing without your yes.

ONE REAL INSIGHT BEFORE YOU PAY · 30-DAY GUARANTEE
Soren: the AI Media-Buying Agent for Meta Ads | Sentrum