Indigo One

What Indigo One is

Indigo One is the AI-native operating layer developed during the Build with Gemini XPRIZE period on top of the pre-existing Indigo software foundation. The XPRIZE project is not presented as if the entire Indigo platform was created during the competition. Instead, the competition work focuses on turning an existing multi-domain business platform into an AI-operated system in which Avery can understand business context, reason with an LLM, select governed capabilities, and produce executable actions through explicit tool contracts.

Indigo itself is a modular business platform with separate backend, admin frontend, and storefront runtimes. Its product domains include catalog, inventory, warehouse, purchasing, orders, payments, POS, people, reporting, delivery, integrations, and other operational surfaces. This foundation provides the business state and deterministic application boundaries that the XPRIZE AI layer can operate against.

Avery: AI as an operating layer

Avery is not implemented as an unrestricted chatbot. Its runtime composes allowed contexts, skills, tool facades, policy, and executable tool contracts. Actual handlers remain behind application boundaries, so model reasoning and business execution are separated.

During the competition period, the AI-native architecture was extended to carry structured client context into Avery, identify semantic application surfaces, and return explicit client actions rather than relying on hard-coded keyword routing. A first operational path used the printing/iPrint workflow as a pilot: Avery gained semantic page routing, a canonical navigation tool, movement policy, and context-aware guidance or navigation depending on the user's instruction. The same architecture is intended to be reusable across other governed business capabilities rather than creating one-off AI shortcuts per screen.

This matters for small businesses because the interface to operational software can become intent-driven without bypassing the controls of the underlying system. The goal is not to replace deterministic business logic with an LLM. The goal is to let AI reason over context and decide which approved capability should be invoked while the application continues to enforce permissions, contracts, validation, and domain ownership.

Gemini integration

Gemini is integrated as a first-class LLM provider in Indigo's backend. The provider supports text generation, streaming, embeddings, system instructions, reasoning metadata, and Gemini function calling. Tool declarations are translated into Gemini function declarations, and function calls returned by Gemini are converted back into Indigo's platform-level tool-call contracts. This allows Gemini to participate in an agent loop instead of being used only for text completion.

The implementation uses the Gemini API through Google's Generative Language endpoint and a dedicated GEMINI_API_KEY execution target. Indigo can also support additional LLM providers, but Gemini has a native provider implementation and can execute at least one LLM call through the deployed application as required by the competition.

Google Cloud runtime

Indigo runs on Google Cloud Run, with application images hosted in Google Artifact Registry and Google Cloud service-account configuration. Google Cloud is part of the production infrastructure used by Indigo, alongside Google services such as Google Maps Platform. The submission evidence separately includes the required Google Cloud billing invoices and Gemini observability records.

What humans do vs. what AI does

Humans define the business, configure the system, own policy and permissions, approve sensitive operations where required, and remain accountable for business outcomes. Avery interprets user intent and application context, reasons about the next permitted action, and can select or propose governed tools. Deterministic application handlers execute those tools and enforce the actual business rules.

This division is deliberate: AI is used where reasoning, interpretation, and orchestration create leverage, while transactional correctness and authorization remain explicit software contracts.

XPRIZE scope and pre-existing resources

Indigo existed before May 19, 2026, and portions of the Avery concept and UI also pre-date the XPRIZE submission period. Those elements must be disclosed as pre-existing resources. The project scope submitted here is the AI-native operational architecture and competition-period implementation built on that foundation. Repository history preserves the distinction between earlier Indigo/Avery work and the AI-native work implemented during the competition period.

The remaining submission evidence—revenue, customer counts, expenses, P&L, customer evidence, Google Cloud invoices, Gemini observability evidence, and the public demo video—should reflect only verifiable competition-period facts.

Built With

• astro
• azure-container-apps
• fastapi
• gemini-api
• google-artifact-registry
• google-cloud-run
• google-maps-platform
• next.js
• postgresql
• python
• supabase
• typescript