Inspiration

Good products often sit unsold because shoppers cannot see how they fit their needs. Merchants respond with broad discounts, while buyers scroll through unrelated clearance listings.

RECLAIM//RUN starts with a different question: what are you trying to accomplish? A creator desk upgrade, a remote-team setup, or a hobby starter kit becomes a shopping mission built from overlooked inventory.

What it does

Buyers describe their goal, set a maximum budget, choose required categories, and compare three bundle strategies: Budget Hunter, Maximum Value, and Merchant Rescue.

Every proposal uses actual workspace inventory. Buyers can inspect products, customize quantities, compare offers in an accessible table, and create a server-calculated quote that reserves stock.

Merchants manage products, import CSV inventory, configure pricing boundaries, approve campaigns, and inspect inventory aging, aggregated mission demand, and payment records.

The optional AI agent uses the buyer's own provider key or a local Ollama connection. API keys remain in memory and are never saved to the database, browser storage, or application logs.

How we built it

The frontend uses React, TypeScript, Vite, custom CSS, and original SVG artwork. AG Grid Community powers the merchant inventory table.

An Express backend manages authentication, role permissions, validated tools, pricing, and payment operations. SQLite provides persistent records and transactional inventory reservations.

The deterministic offer engine calculates prices in integer cents and enforces merchant discount limits, price floors, gross-margin requirements, buyer budgets, and available quantities.

The PayPal sandbox integration implements server-side order creation, PayPal-hosted buyer approval, capture verification, reconciliation, webhook-signature verification, and merchant-authorized full refunds.

Challenges we faced

The central challenge was keeping recommendations separate from financial authority. Model-generated text cannot change prices, invent stock, or approve payments.

We also addressed competing inventory reservations, expired quotes, ambiguous payment outcomes, and duplicate capture callbacks. A browser redirect alone never marks an order as paid.

Accomplishments we're proud of

We built a cohesive buyer and merchant experience with persistent inventory, constrained bundle generation, transactional CSV imports, stock-backed quotes, and clear payment boundaries.

The project passes 39 automated server, pricing, and security tests and five desktop/mobile browser journeys. The production build and GitHub CI pass, and the dependency audit reports zero vulnerabilities.

What we learned

Useful commerce agents need narrow tools, validated inputs, and deterministic enforcement. Giving users control over their provider connection also avoids requiring a shared, platform-funded AI service.

Trust depends on accurately distinguishing recommendations, reservations, approvals, and confirmed financial outcomes.

Current verification status

The initial catalog is explicitly synthetic demonstration inventory. Automated payment and model tests use mocked provider responses.

Live PayPal transactions and genuine runtime AI inference remain unverified because sandbox merchant credentials and a user model connection were unavailable. The application and demo disclose these boundaries rather than displaying simulated success.

What's next

Complete genuine model tool-call and buyer-approved PayPal sandbox verification, then extend the product with production merchant onboarding, tax and shipping calculations, catalog connectors, and operational reconciliation.

The source is publicly available under Apache-2.0, with setup instructions, screenshots, tests, and integration documentation.

Built With

Share this project:

Updates

Submission history