How it works

Four stages, from your URL to approved action.

Learn, work, deliver, approve. Your team learns your business, works it with specialist agents and live data, delivers skimmable briefs into your chat — and anything outbound waits for your explicit approval. That's the whole product.

01
Learn

The team learns your business.

It starts with your URL. Before you've connected anything, the team analyzes your site — value proposition, competitors, positioning — and sends you its first research pass, sources included. Everything it learns from then on lands in a knowledge base you can see and edit.

  • Connect data sources from the connector marketplace — GA4, Search Console, Stripe, Google Ads, Meta Ads, App Store, Google Play and more. Read access first; anything that writes back is a separate, approval-gated step.
  • Upload decks, docs, and PDFs — they're ingested into your team's knowledge base and cited with attribution when used.
  • Your company snapshot — positioning, competitors, target customer — is editable from the dashboard, and every agent reads the updated version.
  • Your workspace runs on its own isolated runtime — your data never shares a process with another customer.
02
Work

One officer, a roster of specialists.

You talk to your Growth Officer. It answers what it can and routes craft work — research, SEO, ads, social, community, customer feedback — to the right specialist, then reports back in the same thread. You never manage the routing yourself.

  • Recurring work comes from the schedule marketplace: a Daily Growth Brief (on by default once you connect a chat channel), competitor watch, customer-feedback digests, weekly deep-dives.
  • Briefs fetch your connector metrics live at send time — the numbers come from your accounts at that moment, not from a cache.
  • If you've set a growth objective, the daily brief tracks the plan: the objective, the current number, this week's plays.
  • Anything the team saves to long-term memory must carry evidence — source, time range, baseline, trigger — enforced at write time.
03
Deliver

Briefs land where you work.

Webchat in your dashboard, or Slack, Telegram, Discord, and Feishu — Slack connects in one click. A brief is a crisp 3–6 sentence writeup built to be skimmed in 30 seconds, not a report you have to analyze.

  • Structured as: headline · what changed · why it matters · the proposed next step.
  • Material claims link to their sources — click through to the competitor's page or the dashboard a number came from. Coverage gaps are flagged, never papered over.
  • Deliverables — audits, drafts, comparison docs — arrive as downloadable files; on Slack and other chat apps they attach natively.
04
Approve

Outbound waits for your approval.

Anything that leaves the building — a social post, a cold email, ad spend, a CMS publish — stays a draft until you approve it. You approve from chat via a signed approval link; the team executes only after your tap.

  • Approvals follow a state machine: proposed → pending_approval → approved → executed (or rejected, or expired after 24 hours).
  • Only an approved, unexpired approval token ever reaches the execution layer — by design, not by configuration.
  • Every approval is auditable: which specialist proposed it, the exact action and its parameters, when it moved through each state, and what came back when it executed.
  • One deliberate exception: reversible micro-engagements (a like, a follow) run without a gate — capped per day and audit-logged.

That's the shape of it — now see it on your own site.

Everything below is how it holds up: the memory that makes "changed" possible, the rule every saved fact has to pass, and a real brief. Read on, or start now.

Start free trial
Memory architecture

The structured state that makes the briefs trustworthy.

Naive "agent with a long context window" setups remember everything and mean nothing — they drift into noise. Your team's memory has two layers instead: shared workspace files every agent reads, and a knowledge base every fact has to earn its way into.

USER.md

The company brief every agent reads

Generated at onboarding from your website and your inputs — company, industry, target customer, competitors, channels. Shared by every agent on your team, read at each session start, and editable by you from the dashboard. Size-capped and enforced at write time.

# Acme · company brief industry: restaurant SaaS competitors: SevenRooms, Tock, Resy target customer: indie restaurant owners
MEMORY.md

Named sections, per agent

  • Last observations —— one line per competitor, keyword, or campaign. The baseline the next diff compares against.
  • Run log —— one line per run, with a 1–5 self-score. How drift gets caught.
  • Durable facts —— curated truths that stick. Pin "Competitor X was acquired in Q3" or "don't monitor company Y" and it stays.
  • Bootstrap —— first-boot rules so a freshly provisioned agent doesn't cold-start blind.
Knowledge base

Facts that carry their receipts

Everything saved for the long term — from your uploads or from a run — goes through one write path that requires an evidence object: source, time range, baseline, trigger. Your team searches it as it works and cites what it uses. You can browse and edit it from the dashboard.

Isolation

No cross-customer memory

Each workspace runs on its own isolated runtime. Workspace files and the knowledge base live per customer, per project. Nothing your agents learn about your business informs any other customer's agents — enforced at the runtime boundary, not just by policy.

Why it matters

Memory is what makes "changed" possible

Every brief's baseline comes from what the team recorded last time. That's what lets a finding say "something changed" instead of "something exists" — and why week two is quieter and sharper than week one.

The evidence discipline

Four fields, no exceptions —
every saved fact carries them.

This is the contract. Anything the team writes to its long-term memory must carry all four fields — the write is rejected otherwise. You never read them as a form, though: briefs cite the source and the period in plain prose, and the structured fields stay underneath, where the knowledge base is searched and audited.

Source
Required
sevenrooms.com/features/waitlist

A URL, file, or API response you can open and verify. Never "my training data" or "general knowledge."

Time range
Required
2026-04-13 → 2026-04-20

The window the fact covers — so you know whether it's this-week news or a last-quarter trend.

Baseline
Required
Feature didn't exist as of 2026-04-12

What the team knew before. A finding without a baseline is just an observation.

Trigger reason
Required
Page changed. Diff: +waitlist module, +FAQ section, +demo CTA.

The specific change that lifted this above noise. Review is fast because you don't have to do the comparison yourself.

A real briefing

Here's what lands
in your Slack.

# growth
M
Market Research LeadAPP · Mon 8:02 AM
SevenRooms shipped a waitlist feature — worth a positioning response this week.
I re-checked sevenrooms.com/features/waitlist against what we had on file from Apr 12, and it changed between Apr 13 and Apr 20: a new waitlist module, an FAQ section, and a demo CTA. No announcement post yet — they're warming it up before launch. This lands squarely on our "no-wait host" messaging. I can draft a comparison post — the publish would go out only after your approval.
JL
Jake8:07 AM
Yes — draft it. I'll review before anything goes out.
M
Market Research LeadAPP · 8:09 AM
On it. The draft "SevenRooms vs Acme — waitlist comparison" will be in your Files panel shortly. Publishing stays gated — when it's ready, you'll get a signed approval link:
Approve publish →Signed link · one action · expires in 24h
Good questions

A few things worth spelling out.

What if an agent gets something wrong?

Tell it — in the same thread. Corrections update the team's memory, so the fix sticks. If something looks systematically off, flag it from the feedback page in your dashboard and a human on our team reviews it with you.

Can I change what runs on a schedule?

Yes. Recurring work comes from schedule templates you install and remove from the dashboard — the Daily Growth Brief, competitor watch, feedback digests, weekly deep-dives. You pick the delivery channel when you install one, and you can trigger any of them on demand with "Run now".

What happens in the first hour?

You paste your URL, your workspace is provisioned — usually in under a minute — and the team sends its first research pass shortly after: a competitor snapshot or site audit, sources included. Connect GA4 or Search Console and your briefs start carrying your real numbers. By week two the baselines are in place, so briefs become change-only.

Where do approvals live after they execute?

Every approval is recorded — which specialist proposed it, the exact action and its parameters, when it moved through each state, and what came back when it executed. Rejected and expired approvals are kept too. If you delete your account, your workspace and its data are torn down entirely.

Ready to see the loop on your business?

Paste a URL and your AI growth team gets to work today.

Start free trialSee pricing
How AgentCeres Works — Your AI Growth Team, Step by Step