SelfmerceBeta
Talk to us

Agents

One chat box. Four standing commerce roles.

Not dozens of independent bots: one copilot surface, four managing agents and a deterministic tool/workflow library. Same state, same limits, same record.

See how it works Trust model

The four managing agents

Each owns an outcome; none of them acts freely.

Studio Director

Turns product and brand context into a catalog and content that are ready to sell.

  • Brief
  • Enrichment
  • Content
  • Brand check
  • Store draft
First mode
A1-A2
Now
Briefs, drafts, enrichment, quality and brand checks.
Long term
Continuously planning and optimising content and store experiments.
Stays human
External publishing, core brand messaging, price and regulatory claims always come for approval.
Product page

Launch Pilot

Checks readiness, coordinates testing and assembles the evidence for go-live.

  • Readiness
  • Setup
  • Test shopper
  • Go / no-go
  • Monitoring
First mode
A0-A2
Now
Readiness, checklist, setup drafts, test orders and go-live evidence.
Long term
Coordinating a cross-channel launch and managing rollback.
Stays human
Domain cutover, PSP/risk policy, tax decisions and go-live stay human-owned.
Product page

Operations Conductor

Runs orders, stock, shipping, support and returns on one shared priority and SLA.

  • Brief
  • Exceptions
  • Routing
  • Recovery
  • Reconciliation
First mode
A0-A2
Now
Daily brief, exception diagnosis, task and action drafts.
Long term
Continuously coordinating orders, stock, fulfillment, shipping, support and returns.
Stays human
Refunds, inventory movement, reroutes, reships and customer goodwill are approved.
Product page
Coming soon

Profit & Growth Controller

Finds profit leaks and drafts the retention, merchandising and experiment plan.

  • Starts with Growth Lite
  • Execution comes later
First mode
A0-A1
Now
KPI explanation, profit leaks and experiment proposals (Growth Lite).
Long term
Optimising retention, merchandising, media and stock against one shared goal.
Stays human
Budget, publishing and every customer-visible change come for approval; execution opens later.
Product page

How do they work together?

The agents share the outcome, not the tools.

What Studio learns isn't rebuilt in Launch; what Operate decides is written to the record Growth reads.

  1. Merchant goals and policyYou define the outcome, the risk appetite and the approval thresholds.
  2. One copilot surfaceOne chat box and a control tower, routing the goal to the right agent.
  3. Four managing agentsStudio Director, Launch Pilot, Operations Conductor, Growth Controller.
  4. Authorised tool and workflow registryAgents only call registered, authorised and typed tools.
  5. Policy, approval and riskEvery action gets a control mode from its risk class and maturity.
  6. Action ledger and evaluationProposal, decision, tool call, result, failure and rollback in one record.

Let's pick the first agent together

Decide which outcome you want to own, and we'll start the agent on that scope.

First pilots start with one product and one agent: Studio Director, Launch Pilot or Operations Conductor.

Talk to us How it works