toolsDetail.backToAllTools
toolsDetail.projectRecordtoolsDetail.categories.aiDevtoolsDetail.statuses.openSourcetoolsDetail.updatedOn

Titan Inventory Agent

A multi-agent inventory synchronization system built for the Google Kaggle Capstone. Ingests and matches CSV hardware listings for expert-mode compatible vehicles.

Titan Inventory Agent is the storefront half of Project Titan, a multi-agent automotive commerce system built for a Google Kaggle vibecoding capstone. It runs live CSV lookups against a mocked inventory server to verify stock for performance-tyre hardware, then negotiates and settles orders directly with a client agent over an autonomous agent-to-agent protocol.

toolsDetail.demo

toolsDetail.overview

Titan Inventory Agent (published as "UHP Tire Storefront Agent — Website A") is a mock e-commerce storefront built to manage specialized Ultra-High-Performance (UHP) tire inventory and autonomously process orders placed by other agents rather than by a human clicking through a checkout flow. It's one piece of a larger submission, Project Titan, which frames itself as an AI agent-driven automotive ecosystem spanning e-commerce, vehicle telemetry, and logistics.

The agent's core job is inventory synchronization: it exposes an Agent-to-Agent (A2A) server that other agents can query and negotiate with over structured JSON-RPC, backed by live CSV lookups against a mocked Google Sheets MCP server. Those lookups check availability for a specific set of "Connect Ready" hardware — Pilot Sport CUP2 Connect tyres, a standalone Central Connection Box SKU, and a four-unit Tyre Sensor hardware kit — and the project's requirements document describes product discovery spanning more than 270 "Expert Mode" compatible vehicles, with pricing calculated in Thai Baht including regional tax and delivery fees.

Ordering is negotiated, not just transacted: incoming JSON-RPC requests go through stock verification, price matching, and dynamic counter-offers using an explicit Input-Required state, and the storefront agent automatically restricts a transaction if the mocked inventory reports a stockout on any component of a kit. Payment is handled through the Agent Payments Protocol (AP2), where a client-side "Concierge" agent signs a digital promissory note that the storefront agent verifies before releasing stock.

In the wider Project Titan design, this storefront agent is meant to sit alongside a Concierge Agent — described in the project's documentation as a "Virtual Tyre Engineer" that watches tire telemetry and notifies the storefront over A2A when wear crosses a threshold — and a dealer-facing logistics role (Aurora) for routing and fitting appointments. This repository is the storefront side of that system; the demo video linked here is shared across both capstone submissions.

toolsDetail.keyFeatures

Agent-to-Agent order negotiation

Exposes an A2A server with a standardized Agent Card and processes structured JSON-RPC requests end to end — stock verification, price matching, and dynamic counter-offers via an explicit Input-Required state — rather than a fixed-price checkout.

Live CSV inventory lookups

Connects to a mocked Google Sheets MCP server to run live CSV lookups against "Connect Ready" hardware SKUs (Pilot Sport CUP2 Connect tyres, a Central Connection Box, and a four-unit Tyre Sensor kit), and automatically blocks a transaction if any component is out of stock.

AP2 payment verification

Integrates the Agent Payments Protocol to verify digital promissory notes signed by a client Concierge agent using RSA-2048 asymmetric cryptography, with replay-attack protection via UUID transaction IDs, millisecond timestamps, and a strict five-minute validity window.

Vehicle-compatible hardware matching

Matches inventory against a catalog spanning more than 270 "Expert Mode" compatible vehicles, with totals — including regional tax and delivery fees — calculated in Thai Baht.

A2UI scheduling and logistics estimates

Renders Google Calendar scheduling for mobile tyre-fitting appointments (via Firebase Auth, inside a secured iframe) and real-time distance, traffic, and delivery estimates via Gemini Maps Grounding, each delivered as a sandboxed A2UI component.

Deterministic and agent-graded evals

Ships a deterministic test suite (test-security.js) that checks signature validity, replay rejection, expired-signature blocking, and input parameter bounds, alongside documented LLM-graded evals for the Concierge agent's anomaly sensitivity and Maps Grounding accuracy.

toolsDetail.architecture

A2A serverAn Express.js (Node.js) backend exposes agent capabilities as a standardized Agent Card and handles JSON-RPC order negotiation and state management; the frontend is React with Tailwind CSS and Framer Motion, built with Vite and ESBuild.
Inventory layerA mocked private MCP server backed by Google Sheets performs the live CSV lookups used for real-time stock verification of Connect Ready hardware SKUs.
Payments (AP2)Order details are signed by the Concierge agent with an RSA-2048 private key and verified by the storefront agent against a corresponding public key, with server-side caching of used transaction IDs to block replay of stale or reused signatures.
A2UI surfacesCalendar scheduling (Google Calendar via Firebase Auth) and logistics estimates (Gemini Maps Grounding via the Gemini API) are each rendered as sandboxed, iframe-wrapped UI components rather than raw agent output.
Multi-agent contextPositioned within Project Titan's broader multi-agent design as the Storefront Agent, coordinating over A2A with a separate Concierge Agent ("Virtual Tyre Engineer") and, per the project's design documentation, a dealer-facing logistics role for routing and fitting-appointment orchestration.
TypeScriptReactTailwind CSSFramer MotionExpress.jsViteFirebase AuthGemini APIGoogle Calendar APIModel Context Protocol (MCP)

toolsDetail.trainingProgramContext

Titan Inventory Agent was built for a Google Kaggle vibecoding agents capstone competition, submitted as the storefront half of a larger project called Project Titan. The storefront agent was entered under the competition's "Agents for Business" track, alongside a companion Concierge Agent entered under a personal-assistant track — the two are designed to communicate with each other over A2A rather than being judged as standalone tools.

The project's own requirements documentation frames the business case around inventory and order-processing automation, citing a target of reducing a stated annual operating-expenditure baseline by roughly 11 percent through agentic automation of the e-commerce lifecycle. That figure comes from the project's planning documents rather than an independently verified result, but it reflects the cost-automation angle the submission was built to demonstrate — alongside protocol-level building blocks like A2A negotiation, AP2 payment verification, and MCP-backed inventory sync that are broadly reusable outside the tyre-commerce scenario.

toolsDetail.bottomCtaQuestion

toolsDetail.browseAllTools