Back to work

Clopi

An assistant that answers a small business's customers from the business's own facts.

Category
AI Customer Service SaaS
Status
Concept
Technologies
LLM APIs, AI Agents, MCP, Next.js, TypeScript, PostgreSQL
Index
04 / 05
Asset pending

Conversation view with the answer traced back to a knowledge record.

public/projects/clopi/cover.webp

01Overview

Clopi is a concept for a customer-facing assistant aimed at UMKM — micro, small and medium businesses — where the owner is also the support desk.

02Problem

For a small business, customer questions arrive continuously and most of them repeat: is it in stock, how much, when does it ship, are you open. The cost is not difficulty, it is interruption — the same five answers given a hundred times, across several channels, at hours the owner is not working.

03Approach

Give the business a small, owned knowledge base, and put a language model in front of it as an interpreter rather than an author. The model's job is to understand the question and find the right record; the answer itself comes from facts the owner controls and can correct.

04System

  1. CustomerAsks in the channel they already use.
  2. AssistantInterprets intent, retrieves an answer.
  3. KnowledgeThe business's own stock, pricing, policy.
  4. OwnerReceives only what needs a human.
Knowledge storeProducts, pricing, policies and hours as structured records the owner edits directly — the single place an answer can come from.
RetrievalIncoming questions are resolved against that store before generation, so an unknown answer stays unknown instead of becoming plausible.
EscalationAnything outside the store's coverage is handed to the owner with the conversation attached, rather than guessed at.
Channel adaptersThe assistant meets customers where they already are, so the business is not asking anyone to learn a new surface.

05Interface

Asset pending

The owner's knowledge store — products, pricing, policy, hours.

public/projects/clopi/knowledge.webp