InboxSherpa: A Privacy-First AI Email Assistant for Small Shops

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle

What I Built

InboxSherpa is a privacy-focused AI assistant for small shops and solo sellers who spend too much time answering customer support emails.

Paste a customer email and your shop's policy, and InboxSherpa classifies the message as Shipping, Refund, or General, summarizes the issue, and drafts a reply based on the provided policy. The reply can be edited and copied before sending. It doesn't send emails automatically.

It's designed for small businesses that want to spend less time on repetitive support replies without handing customer messages to a third-party AI API.

Demo

Code

Aman-Mondal / InboxSherpa

InboxSherpa

A privacy-first AI assistant for customer-support email triage.

InboxSherpa helps small businesses make sense of incoming customer emails and draft replies based on their own shop policies. It classifies a message, summarizes the customer's issue, and generates an editable response for a human to review.

Project status: MVP / Hacktoberfest Side Hustle challenge
AI model: Gemma 3 1B via Ollama
Current mode: Local inference on the machine running the backend

The problem

Small shops and solo sellers often spend time answering repetitive questions about shipping, refunds, and general support. InboxSherpa aims to reduce that busywork without automatically sending messages or taking decisions away from the business owner.

Features

• Email triage: Categorizes messages as Shipping, Refund, or General.
• Issue summary: Produces a short description of what the customer needs.
• Policy-aware drafts: Uses the shop policies provided by the user to draft a reply.
• Editable responses: Review and edit the generated…

View on GitHub

How I Built It

• Frontend: React, Vite, CSS, and Lucide icons
• Backend: Node.js and Express
• AI: Gemma 3 1B through Ollama, running locally

The frontend sends the email and shop policy to the Express API. The backend passes them to the local Ollama model and returns a category, summary, and suggested reply.

The model is instructed to use the supplied policy and avoid inventing details. The user can review and edit the draft before using it.

Why Does Open Innovation Matter?

Using an open-weight model with local inference makes it possible to experiment without depending on a paid third-party AI API for every request. It also gives the shop owner more control over where customer email content is processed, since the current setup uses a local Ollama instance.

This is an early prototype, not a complete support platform. Local inference depends on the machine running the model, and drafts should always be checked before sending.