Platform · Agents and AI

Agents as teammates. On your live operational data.

Not a chatbot bolted on the side. CCEN agents do real work on your real data. They draft real POs, place real holds, file real tickets. They act under your team's identity, write to the audit log, and ask for approval when policy says to. Like a smart junior teammate, not a sparkle effect.

PO Planner
acting as sarah@miabellebaby.com · audited
Connected
Plan the next PO with our top vendor. Reorder anything below 6 weeks of cover before the Mother's Day pull-forward.
Pulling sell-through from last 14 days, weighted by Marketing Moments calendar.
reports.runok
sell_through_by_sku, window=14d, weight=marketing_calendar
48 SKUs returned
inventory.weeks_of_coverok
vendor=babynest_co, threshold=6
14 SKUs under 6w
supply.draft_poready
vendor=babynest_co, items=14, target_cover=10w
Draft PO-2451 created
Approval required
PO-2451 · BabyNest Co · $48,210.00
14 SKUs · Net-30 · target arrival 2026-05-12
policy: dual sign-off >$25k
The thesis

AI is a surface. Not a skin.

Most "AI features" in commerce platforms are a chat panel wrapped around a system prompt. The model sees a synthetic sample of your data, calls a generic public API, and returns advice your ops lead has to reverse-engineer back into the real tables. The agent never actually does the thing. It describes the thing. The seam between the agent and the work is exactly where the value used to be.

CCEN agents are different in two ways. Technically, they actually do the work. They read and write your real orders, products and customer records through the same controls your team uses. Organizationally, they act under a real operator's identity, see only what that operator can see, and route to that operator's approval flows. Every action is logged.

The result is closer to a junior teammate than a chatbot. The PO planner drafts a real PO. The fraud sentinel places a real hold. The CS triage agent assigns a real ticket. None of them ship anything to a customer or a vendor without an explicit human approval where the policy demands one. The agent surface is a worker shelf, not a sparkle effect on the dashboard.

What makes a CCEN agent different

Six properties. Load-bearing.

Real data, not a sample

Agents read and write the same orders, the same inventory, the same customers your team uses every day. No sandbox, no demo mode pretending to be production, no synthetic snapshot.

Every action logged

Every action by an agent is logged: who asked, which agent did it, what prompt produced it. You can undo any single action, replay an entire session, or export the whole log.

Forkable apps under the hood

Agents lean on your apps. Customize an app (fork it, or have AI customize it for you) and your agents pick up the changes automatically.

Drawer-based UX

Agents live in a side drawer that follows you across the app. Pin one to your screen, stack two, summon a fresh one with a keystroke. The drawer carries the context of the screen you are looking at.

Smarter when you install apps

When you install a new app, your agents notice. Install a forecasting app today, and tomorrow your PO planner uses better forecasts. No code changes.

Approval flows in plain English

You set the rules in plain English: 'Two people sign off above $25k. One person for transfers. No approval for read-only questions.' The agent routes to whoever needs to approve, and waits.

The agent runtime

From prompt to PO sent.

One prompt. Two reports run. One inventory check. One PO drafted. One approval routed. Each step is a real tool call against real tables, each result is auditable, each write requires the policy-defined approval.

  • The agent acts as you. Every action is logged under your name.
  • You see every step as it happens, not a final summary you can't audit.
  • Approve, edit the line items yourself, or reject. All three are one click.
  • Resume the session from any device, any time.
PO Planner agent
session 8f12 · resumable · 4 tool calls
Plan the next PO for BabyNest Co. Reorder anything below 6 weeks of cover, factor in the Mother's Day pull-forward next month, target 10 weeks of cover after receipt.
Pulling sell-through from last 14 days, weighted by the Marketing Moments calendar. Mother's Day is tagged as a +30% lift event for sleepwear and bath SKUs.
reports.runok
{ "report": "sell_through_by_sku", "window": "14d", "weighted_by": "marketing_calendar" }
48 SKUs returned · sleepwear and bath SKUs flagged for +30% lift
inventory.weeks_of_coverok
{ "vendor": "babynest_co", "threshold": 6 }
14 SKUs below 6w cover · 9 are sleepwear, 5 are bath
Drafting a PO targeting 10 weeks of cover post-receipt. Receipt assumed in 21 days based on BabyNest's last 6 inbound shipments.
supply.draft_pook
{ "vendor": "babynest_co", "items": 14, "target_cover_weeks": 10, "lead_time_days": 21 }
Draft PO-2451 created · $48,210.00 · 14 line items
Validated against vendor minimums. PO is above the policy threshold for dual sign-off ($25k). Routing to Marie (CFO) for second approval after yours.
Approval required · step 1 of 2
PO-2451 · BabyNest Co · $48,210.00
14 SKUs · Net-30 · target arrival 2026-05-12
policy: dual sign-off
Ask a follow-up. Or pin this drawer to your screen.
Built-in agents

Six first-party agents. Day-one useful.

Every CCEN tenant ships with these. Each one names what it can do and what it cannot do without you. The line between the two is the approval policy, and the policy is yours to set.

PO planner

Drafts purchase orders weighted against the Marketing Moments calendar.

What it can do
  • Run sell-through reports across windows
  • Draft POs targeting weeks-of-cover
  • Route to dual sign-off on policy
What it cannot do without you

Send a PO to a vendor without explicit operator approval.

Inventory rebalancer

Watches stock positions across locations and proposes transfers.

What it can do
  • Detect SKUs under reorder point per location
  • Draft transfer orders with lane economics
  • Flag SKUs at risk of stockout before campaigns
What it cannot do without you

Commit a transfer that crosses a tax-jurisdiction boundary without approval.

CS triage

Reads incoming tickets, attaches order context, drafts replies in your brand voice.

What it can do
  • Classify and route tickets to the right queue
  • Draft replies grounded in real order history
  • Escalate SLA breaches to a human agent
What it cannot do without you

Issue a refund or RMA without operator approval.

Fraud sentinel

Scores every order at checkout against your historical ground truth.

What it can do
  • Hold orders that match learned fraud patterns
  • Cross-check billing and shipping mismatch
  • Surface chargeback risk on the order detail
What it cannot do without you

Approve a held order back into shipping without operator review.

Marketing optimizer

Watches MTA and MMM signals, suggests budget shifts across channels.

What it can do
  • Reweight channel spend based on attributed revenue
  • Flag ads with slipping creative fatigue
  • Plan promo cadence around Marketing Moments
What it cannot do without you

Push a budget change to a connected ad platform without approval.

Cashflow forecaster

Projects cash position 90 days out using inbound POs, payouts, and chargeback rates.

What it can do
  • Build a daily cash forecast from live ledger data
  • Flag covenant risk on a credit facility
  • Propose payment-timing changes on open A/P
What it cannot do without you

Initiate a payment or wire without operator and finance approval.

Apps make agents better

Install an app. Your agents get smarter.

Every app you install adds new tools your agents can use. Install a forecasting app today, and tomorrow your PO planner uses better forecasts. Install a tax app, and your finance agent files cleaner. The marketplace makes your agents better, automatically. No code changes.

Fork an app and add your own custom tools, and your agents pick those up too, scoped to your team.

For developers: how tool registration works
Install an app · agents notice
  • Forecasting app
    Your PO planner gets sharper forecasts on the next prompt.
  • Tax compliance app
    Your finance agent files cleaner returns automatically.
  • B2B portal app
    Your CS triage agent gets wholesale-aware ticket routing.
When apps are installed, agents discover what they can do and start calling those tools, scoped to your permissions and approval rules. From your seat, the assistant just gets better the day after a new install.
Custom agents

Build your own agent in four steps.

01

Define the job

Name the agent, describe what you want it to do, in plain English. Read-only by default. You explicitly grant write access where you want it.

02

Pick which apps it can use

Inventory, Shipping, Finance, Marketing. The agent uses whatever's installed. Swap apps later and your agent keeps working.

03

Set the approval rules

Two people for big POs, one for transfers, none for questions. Your policy, your call. Logged automatically.

04

Ship to your team

Decide who can use it. One workspace, one team, the whole company. It shows up in the agent shelf and works the same as the built-in agents.

The PO planner cut our weekly buying meeting from three hours to twenty minutes. We still approve every line. We just stopped spending the morning building the spreadsheet that built the PO.
DM
Daniel Mendoza
Head of Operations · Floor & Forge
Agents and AI

See an agent draft a real PO on your data.

Bring 30 minutes and a sample of your last week of orders. We will set up the PO planner, the rebalancer, and the CS triage agent on your tenant and watch them work.