YogentMerchant AI Agent
What it does Why not a chatbot Product Traction

Multi-channel e-commerce · TikTok Shop · Amazon · Temu

AI agents that run the busywork of multi-channel e-commerce

Product research, competitor tracking, daily reporting, listings, ads, videos and comments — the procedural work that today scales only by hiring. Merchant Agent turns each of those into a role an agent holds, working the seller centers directly, in the seller's own browser.

What the agents do Why not just use Claude
5
stores in the pilot
240h
saved per merchant each week
~90%
reported cut in operations cost

What it does

Four jobs a seller stops doing by hand

Each one is a standing role with its own tools, skills and memory — not a prompt someone retypes. They run on a schedule and escalate to a person when judgement is required.

Scenario 01

Which product should we sell next?

  • Collects TikTok and Temu best-seller leaderboards daily, in the seller's own Chrome
  • Tracks price, GMV, daily sales, creators, videos and commission per product
  • Flags follow-sell fit and IP risk before anyone spends on stock
  • One click copies a winner into My products; the rival shop goes to Competitors
Scenario 02

A rival cut its price overnight. Did we notice?

By hand, a dozen rival listings get checked a few times a week — price cuts are noticed days late, after the sales are already lost.

  • Visits chosen rival shops every day in an isolated browser profile
  • Logs price changes, promotions, new SKUs and review swings
  • Spots their best-performing videos and creators
  • Suggests a response: match price, bundle, or push our own video
Scenario 03

An 11-page report, ready at 4 a.m.

Built from seven seller-center pages: orders, finance, returns, reviews, shop health, analytics and ads.

  • Profit per SKU after shipping, fees, ads and cost of goods — losers in red
  • Refunds with time left to act, and new 1–3 star reviews
  • Shop health score and violations, with a warning before the score drops
  • Today's to-dos: must-fix items in red, suggestions in green
Scenario 04

Videos and comments, from an iPhone

  • Publishing agents post product videos to TikTok and Instagram from a phone, as a person would
  • Replies answer buying questions using facts from My products, so nothing is invented
  • Works the first hour after posting: pin good comments, reply fast, start a discussion
  • Every reply is drafted for the seller to approve in a batch; complaints and refunds go to a person

TikTok and Instagram engagement agents are built; live-account pilots are next.

Competition

Why not just use Claude or Codex?

Models like Claude are the engine. Merchant Agent is the workbench sellers actually use.

General AI agents todayMerchant Agent
Every task is a back-and-forth chat, and the right window is hard to find Tasks run as scheduled workflows; results land in the Inbox and Dashboard
Sessions are grouped by conversation, not by agent Each agent has one home: its tasks, history and output together
Sessions don't map to the roles a store actually has Agents mirror real roles: daily reporter, ads optimizer, notifier
A chat window can't show the many signals an e-commerce call needs Screens built for e-commerce: shops, SKUs, ads, profit and to-dos side by side

Product

Built for how multi-shop sellers actually work

  • Browser control. Many seller-center tasks have no API at all. The agent works the web pages directly, inside sessions that are already signed in.
  • IP and profile isolation. Shops sharing a device or IP risk being linked. Each shop runs in its own browser profile and IP.
  • Posting from a phone. Videos and comments go out from an iPhone, the way sellers already do it.
  • Roles accumulate. Every correction becomes a new version of that role, so the asset compounds per store instead of resetting each session.

Automation

The morning routine finishes before anyone sits down

Agents are chained into workflows that run on a schedule: pull the data, write the report, post it to the warehouse group — done. Risky steps wait for approval, and anything that fails lands in the Inbox with its context instead of disappearing.

The Merchant Agent Hub lets a seller install a role that already works for someone else, rather than building every agent from scratch.

Shops

Every store, one place

All stores on all platforms, with each agent scoped to what it may touch.

My products

One catalog

Published in each platform's own format.

Dashboard

Real profit per SKU

After shipping, fees and ads — not gross revenue.

Inbox

The human queue

Refunds, bad reviews, failed labels and approvals as to-dos.

Research

Daily best-sellers

TikTok and Temu leaderboards, with IP risk flagged.

Competitors

Rival tracking

Listings and prices watched automatically.

Videos

Short video without an editor

Scripts, voiceover, subtitles and finished cuts.

Chats

Ask in plain words

"Why did ROI drop yesterday?" — answered from the store's own data.

My Agents

Role-based agents

Daily reporter, ads optimizer, notifier, and the rest.

Workflows

Scheduled runs

Chained agents that finish on their own.

Agent Hub

Ready-made roles

Install agents and workflows in one click.

Mobile

From your phone

Check results, give feedback, send product videos.

Traction

Merchants are already running on it

  • 5 stores in the pilot, onboarded directly by us.
  • 240 hours saved per merchant each week.
  • Live on our own family's stores since 2026 — we are the first user, every day.
“Free hands and cut 90% cost on operations, made us focus on product iterations.” — pilot merchant

Market

~11M stores to serve

US marketplaces are run by sellers in China and the US: ~1.65M active Amazon sellers, 215K+ on TikTok Shop and growing 25% a year, ~300K Temu vendors. Counting roughly five stores per seller puts the serviceable base near 11 million stores.

Sources: Marketplace Pulse (Jan 2026), Modern Retail (May 2026), Chinesellers (Nov 2024). Stores per seller and price per store are our own assumptions, not observed figures.

Seeing it

The honest version of this is a screen share

Slides flatten the part that matters. What is worth looking at is the running operation: the agents that went out this morning, the runs that finished, and the ones sitting in the queue because they needed a person. We would rather show that than describe it.