Built for Heartland Merchants · SG & SEA

The always-on
AI shopkeeper's assistant
for the heartland

AgnesClaw runs 24/7 inside your shop — drafting replies to reviews and messages, reading footfall and demand, flagging supply trouble before it hits, and keeping your storefront visible online. You just talk to it. It remembers everything.

Say: "Agnes, check my reviews"
New 5★ review Restock kaya soon Footfall up 12%
The Reality on the Ground

One pair of hands.
A counter you can't
leave.

The neighbourhood uncle, the mini-mart auntie, the kopitiam stall — they run on thin margins and zero spare time. There is no marketing team, no social-media person, no analyst.

So reviews go unanswered, WhatsApp messages pile up, the Google Business photos are three years old, and stock decisions are made on gut feel. Every hour spent on a phone is an hour away from the customer in front of you.

Tied to the Counter
Can't leave the shop to restock, chase suppliers, or update listings
Slow Replies, Lost Trust
WhatsApp orders and reviews wait hours — customers move on
Invisible Online
Stale photos, thin presence — new neighbours never find the shop
Pressures AgnesClaw takes off your plate
Review Response Speed
Businesses that reply to reviews see roughly +12% more reviews and higher ratings — yet most heartland shops never reply at all.
Message_Agent drafts a warm, on-brand reply to every Google, Facebook and Instagram review within minutes — you just approve by voice.
Reply Latency
Around 9 in 10 customers expect a reply within minutes. A reply that arrives the next day usually arrives too late to win the order.
AgnesClaw watches your inbox round the clock and pre-writes responses to WhatsApp and Telegram enquiries the instant they land.
Outdated Storefront Photos
Listings with fresh photos get far more clicks and direction requests — but stall photos are often years out of date.
Marketing_Agent generates fresh posters, product images and short clips on demand, so the storefront always looks open and alive.
Stock & Demand Guesswork
Gut-feel ordering leaves shelves empty on the days that matter and overstocked the rest — both quietly eat thin margins.
Trend_Engine reads footfall, weather and local demand signals to suggest exactly what to reorder, and when, before you run short.
No Crisis Playbook
One viral bad review or a sudden supply gap can sink a week of takings — and there's rarely a plan ready when it hits.
Complication_Advisor runs tested protocols for viral reviews, supply disruption, inspection failures, staff shortages and equipment breakdowns.
How It Works

Speak. Route. Done.

PIPELINE
Voice to live shop actions
Hover any step below to trace its path through the live pipeline — from a spoken request to the specialist that handles it.
FEATURE CARDS UPDATE MERCHANT VOICE GEMINI LIVE API NATIVE AUDIO · BARGE-IN · hlm_Orchestrator (ADK) MessageAgent MarketingAgent ComplicationAdvisor VisualAssistant TrendEngine
01
Voice_In
Voice In
"Agnes, draft a reply to that 2-star review." Speech streams in as audio — talk over it any time, barge-in just works.
Native audio Barge-in
02
Gemini_Live
Gemini Live
Gemini 2.5 Flash native audio understands the request in context — your shop, your catalogue, your past chats — no transcribe-then-think delay.
Gemini 2.5 Live API
03
ADK_Routing
ADK Multi-Agent Routing
The hlm_Orchestrator hands the request to the right specialist — Message, Marketing, Trend, Visual or Complication — all in one live session.
Google ADK transfer_to_agent
04
Feature_Cards
Cards Update Live
Drafted replies, generated posters, reorder suggestions and sentiment alerts appear on your console — approve, tweak or send, all by voice.
Live UI Voice approve

One orchestrator, five specialists

A single root agent listens, understands and delegates. Each specialist owns one job — and they share persistent memory of your shop, your products and every past conversation.

ROOT AGENT
hlm_Orchestrator
The brain of the shop. Holds Gemini Live's bidirectional audio session, reads intent, and routes every request to the right specialist — while keeping persistent memory of merchant preferences and the product catalogue.
Gemini Live ADK LlmAgent Persistent memory
01
Conversation
Message_Agent
Drafts warm, on-brand replies to customer messages and reviews across WhatsApp and Telegram — ready for you to approve and send.
WhatsApp Telegram Review replies
02
Creative
Marketing_Agent
Generates fresh posters, product images and short promo video on demand via the Agnes API — keeping the storefront looking open and inviting.
Posters Images Video
03
Crisis Protocols
Complication_Advisor
Runs tested playbooks the moment trouble hits: viral negative review, supply disruption, health-inspection failure, staff shortage, equipment breakdown.
Viral review Supply gap Inspection
04
Always Watching
Visual_Assistant
Autonomously monitors social media and the screen — catching new reviews, mentions and posts across Google, Facebook, Instagram and Reddit as they appear.
Social monitor Screen aware Autonomous
05
Analytics
Trend_Engine
Aggregates footfall, sentiment and sales data into plain-language insights — and turns local demand signals into concrete inventory suggestions.
Footfall Sentiment Inventory
MEMORY
Shared Across All
Persistent Memory
Every agent reads and writes the same long-term memory — merchant preferences, product catalogue and past interactions — so context never resets.
Preferences Catalogue History

Built on Gemini Live & Google ADK

A production-grade stack: real-time multimodal AI, purpose-built ADK agents, a streaming WebSocket server, persistent session storage and the messaging APIs heartland shops already use.

Live Demo

Put a tireless shopkeeper's
assistant behind your counter

Connect your microphone, say a command, and watch AgnesClaw route it to the right agent — drafting replies, reading trends and watching your reviews, all in real time.

Click the orb to connect · Say "Agnes, check my reviews" to begin

AgnesClaw is built to demonstrate the capabilities of the Gemini Live API & Google ADK for heartland commerce. Trend figures shown are illustrative. Always review AI-drafted content before publishing or sending to customers.