Features

Multi-channel catalog

One canonical product / offer model maps to Shopify, Amazon SP-API, WooCommerce, BigCommerce. Add a channel, see the same catalog through it. Inventory updates write through to every connected channel. Webhooks come in, get verified, and update canonical state.

• Adapters for Shopify (full sync + draft orders), Amazon SP-API (catalog + orders), WooCommerce (REST + webhooks), BigCommerce (V3 API). Provider-agnostic by design.
• HMAC verification on every webhook. Dedup by delivery id.
• Stripe for payments, EasyPost for shipping labels, Stripe Tax for sales-tax calculation. Each one swappable without touching your data.

Studio — AI content generation

Studio is a pure orchestration layer over LLM and image-generation adapters. Pass it a product + an intent template, get back a canonical ContentAsset with full provenance — model, version, token usage, prompt hash, source product references. Every generation is tagged with cost-attribution so you see the AI spend per merchant per campaign per day.

• LLM adapter wraps the LiteLLM gateway plus direct Anthropic / OpenAI / on-device 4090 fallbacks.
• ImageGen adapter for hero shots, social cards, lifestyle composites — providers swap without code change.
• Provenance carries through to every published platform — you can audit which model wrote any post you publish.

Agent-readiness rubric

Every offer is scored 0–100 against a public rubric (versioned — `hygoshop-v1.0.0` today). The rubric measures how well an offer answers the questions an AI shopping agent would ask. Missing alt-text? Score down. Description under 50 chars? Score down. No GTIN, no SKU, no inventory? Score down. The Studio loop targets lifting this number.

• Score the catalog before you publish; see what's worth fixing first.
• Re-score on every publish; see whether the new content moved the number.
• Bulk re-score against new rubric versions — no manual catalog audit.

Workflow automation — five canonical starters

Pre-wired workflows for the five jobs every merchant runs: new-product launch, agent-audit, abandoned-cart recovery, review-to-content, reorder prediction. Each one is a DAG of adapter calls — Studio, Commerce, Social, Email — that you can customize without authoring nodes from scratch.

• Launch — new product → Studio generates content → claim review (restricted categories route through human review) → push to commerce + social channels.
• Agent-audit — score the offer; if below threshold, hand off to an operator for remediation; re-score after.
• Abandoned-cart — detect inactive cart → Studio generates recovery email → wait for conversion or expire.
• Review-to-content — inbound review → Studio asset → publish across configured platforms.
• Reorder — predict from order history → notify customer → optional auto-create draft order on connected channels.

Operator marketplace

Some decisions shouldn't be automated. Restricted-category content (supplements, health, financial, alcohol) routes through human review before publish. Agent-audit hands off low-readiness offers to operators with the right skills. Reorder workflows can request operator-assisted draft orders for high-touch merchants. Operators are paid per-task via Stripe Connect; escrow holds funds until completion.

• Skill-matched task routing — operators see only tasks matching their declared skills.
• Atomic claim + escrow lifecycle. Held → released | refunded with full audit log.
• Phase J ships invite-only operator onboarding; self-serve opens post-launch.

Multi-tenant isolation

Three isolation tiers, picked per merchant.

• Shared — single Postgres schema with row-level security keyed on merchant_id. Loud-fail mode: a query without merchant context throws, never silently returns zero rows.
• Isolated schema — your data lives in a dedicated Postgres schema, isolated from other merchants at the database level.
• Isolated database — your data lives in a dedicated Postgres database. Migrate tier only.

What's intentionally NOT here

• No on-platform PII storage beyond what's required to operate. Customer emails are hashed; analytics is opt-in.
• No vendor lock-in. Every external service runs through an adapter; switching providers is a configuration change.
• No "coming soon" features in production. If it's documented here, it's wired and testable today.