Running in production · Seoul

Commerce operations that run themselves.

onnubi builds AI-native operations software for multi-channel sellers. Supplier changes, pricing, stock and customer inquiries are handled automatically, with guardrails, and a human is pulled in only when it matters.

See how it works → What we've shipped
ops.onnubi · event streamLIVE REPLAY
Collect
Judge
Guard
Act
Report
0

sales channels synced from one console

0+

supplier price points refreshed per collection run

0

automated tests gating every production deploy

0min

customer-inquiry sweep across every channel

The problem

Small sellers run enterprise-grade complexity by hand.

A single fruit listing can depend on a wholesaler's notice board, a price sheet that changes overnight, four marketplaces with four different APIs, and a customer asking a question at 2 a.m. Miss one signal and you ship at a loss, or oversell stock you don't have.

  • 01
    Signals are scattered

    Sold-out notices, cost changes and policy updates arrive through portals, spreadsheets and chat messages.

  • 02
    Every channel is different

    Naver, Coupang, Toss and 11st each model price, options and discounts their own way.

  • 03
    Mistakes are expensive

    A wrong price or a missed stock-out costs margin, ranking and seller score at once.

Products

One operating layer, three products.

Each one started as a tool for our own stores and runs on real orders before it's offered to anyone else.

Live

onnubi OPS

The operations console behind our own multi-channel stores, including an inventory-free fruit fulfilment business. It watches suppliers, decides what changed, and updates every channel safely.

Supplier notices & cost sheets collected on schedule
Automatic sold-out / restock across channels
Margin-floor pricing with a 3-layer guard
Best-available supplier switch detection
Customer-inquiry watch on every marketplace
Triage-based alerts: urgent, digest, log-only
Naver Smart StoreCoupangToss Shopping11st
ChannelLast syncOptions in sync
Naver12s ago
Coupang41s ago
Toss7s ago
11st26s ago
Pilot

AAF Chat

A helpdesk where staff and customers each write in their own language. Two-way translation, AI reply drafts and thread summaries.

Do you ship to Germany?customer · English
네, 독일 배송 가능합니다.staff · Korean → sent in customer's language
In build

Content Engine

One source turned into on-brand posts for every channel, with product facts pulled from the store, never invented.

1 source
Shopify article
Instagram
Pinterest
How it works

A closed loop from signal to shelf.

Every change follows the same five steps, and every write is backed up before it happens.

01

Collect

Official supplier APIs, spreadsheet exports and notice boards, pulled on schedule or triggered by a new message.

02

Judge

AI reads notices and images and classifies them: sold out, price change, new option, or noise.

03

Guard

Hard rules decide if a change is safe to apply automatically or must wait for a person.

04

Act

Channel APIs are updated through a GET → backup → PUT gate, dry-run by default.

05

Report

Urgent issues alert immediately; everything else rolls into a scheduled digest and an audit log.

Guardrails

Autonomy, with brakes built in.

Automation is only useful if it's trusted. These limits are enforced in code, not left to a prompt.

±30%

Sudden cost swings are held

If a supplier cost jumps or drops beyond the band, the change is parked for review instead of applied.

+20% · 24h

Increases are capped and cooled down

One automatic price increase is limited in size, and the same option can't be raised again within a day.

List price: fixed

Only the discount moves

The original list price is never changed, because changing it hurts marketplace ranking. New listings start at a 40% discount, leaving room to move the selling price either way.

list price ₩50,000 · lockeddiscount 40% → selling ₩30,000
How we build

Founder-led, tested like a platform.

0

automated tests (Node + Python) gate every deploy. The stack spans Cloudflare edge functions, a fixed-IP server proxy, on-prem workers and a React console, built AI-first with Claude Code.

Engineering
AI-first development

Feature work, refactors, test suites and incident fixes are done with AI coding agents, reviewed against tests before shipping.

Judgement
Reads supplier signals in production

AI classifies sold-out and price notices and reads product images that rules alone can't parse.

Ops chat
Runs ops by chat

Free-form operator commands from a phone are turned into safe, scoped actions.

Copy QA
Reviews copy before it ships

Product copy is drafted, cross-checked and verified against source facts before publishing.

Progress

Shipping in public, month by month.

AAF Chat becomes its own product

A store chat feature grows into a standalone multilingual helpdesk with its own backend, database and admin app.

OPS console goes live on the edge

The operations console moves to Cloudflare with a fixed-IP proxy for marketplace APIs and a PIN-gated admin.

Message-triggered supplier collection

Supplier announcements arriving by chat start triggering collection automatically on dedicated hardware.

List-price invariance enforced in code

The pricing engine refuses any change that would touch a list price; all adjustments go through discounts.

onnubi incorporated · official supplier APIs

The company is founded, and collection moves from scraping toward official supplier APIs and exports.

Every channel watched

Customer inquiries are monitored on all four marketplaces through official APIs, with urgent cases routed to a person.

Contact

Contact us.

Partnerships, pilots or questions about how we build. Leave a note and we'll reply by email.