Meet Didi, your AI
store manager
Didi isn't a chatbot. It's a multi-agent system that wakes up every morning, diagnoses your store, and prepares specific owner-approved fixes — powered by Gemini 2.0 and Fivetran MCP.
How a mission works
You tap "Run Didi Diagnosis" or the scheduler fires at 7am.
Fivetran MCP fetches fresh data from Shopify, Meta, GA4, inventory, support, and reviews.
7 specialized agents run in sequence, each contributing evidence to a shared context window.
Gemini 2.0 Flash synthesizes all evidence and returns a structured JSON diagnosis with ranked leaks.
5 specific fixes are drafted — emails, ad pauses, supplier messages, copy edits, support replies.
Nothing happens until you tap Approve. Every action is risk-leveled and shows confidence %.
The agent team
Built with
Structured diagnosis output via responseSchema. Function calling for evidence ranking.
Data freshness checks, connector status, schema inspection across 8 ecommerce sources.
SSR-first React framework with server functions. Zero-latency navigation, edge-compatible.
Every action requires explicit owner approval. Risk-leveled (low/medium/high). Full audit log.
One trusted data layer
Didi queries all data sources through Fivetran's MCP server — providing standardized, fresh, trusted data from Shopify, Meta Ads, Google Ads, GA4, Stripe, Help Scout, inventory files, and review platforms.
Free demo · No signup required