Mission & market
Historical design content: references below to a planned LightRAG/graph backend are superseded by AGNT2-303. Current knowledge uses word/vector retrieval; those references are not implementation or deployment instructions. Knowledge.
Why we build Agentfy.ai
Agentfy.ai's mission: become the AI agent for CMS / e-commerce platforms — starting with OpenCart. Every online store gets a capable, embeddable agent — without running an always-on bot per store.
The whole architecture exists to serve this one goal. Everything else — the ephemeral runtime, the layered slices, LightRAG, the worker — is in service of "an AI agent for every store, cheap enough to give to all of them."
The problem
There are millions of OpenCart (and WooCommerce / PrestaShop / Magento / Shopify…) stores, and almost none have a genuinely useful AI agent. They want one that can:
- answer customer questions 24/7 (support, shipping, returns),
- find products and help choose (catalog Q&A, upsell),
- check order status, help with checkout,
- automate merchant/admin chores (update products, summarize orders).
Building a custom agent per store is expensive, and running a dedicated always-on bot per store doesn't scale — that's exactly the "pay for presence, not work" cost trap. Most stores idle most of the time.
Why our architecture is the unlock
| Need | How Agentfy.ai delivers it |
|---|---|
| Thousands of stores, cheaply | Ephemeral agents — a store's agent is a DB row; it runs only on traffic. Scale by traffic, not store count. (Overview) |
| One platform, many merchants | Multi-tenant — team = the store/merchant; everything scoped per team. |
| Knows the store's catalog | Knowledge via LightRAG — ingest the catalog/docs, ask in natural language. |
| Talks to the actual store | MCP + connectors — attach the OpenCart API (products, orders) as tools; pluggable per store. (Brain, hands & knowledge) |
| Lives on the storefront | Channels — embeddable web widget on the store, plus Telegram/Slack. |
| Does real work | Worker — bash/browser/long tasks (scrape, generate, automate). (Worker) |
| Fair pricing | Usage-based billing — per-store subscription + metered tokens/runtime, not per always-on bot. |
The ephemeral model is the commercial unlock: it makes "an agent for every store" economically possible.
Target market
- OpenCart first — a large, under-served base of self-hosted stores with an extensible API.
- Then other e-commerce CMS — WooCommerce, PrestaShop, Magento, Shopify, etc. Same model: one agent per store, knowledge from the catalog, tools via the platform's API.
- Adjacent: any CMS/SaaS where a per-tenant, knowledge-grounded, tool-using agent adds value.
What a merchant gets
- A branded AI agent on their storefront in minutes — no per-store engineering.
- Grounded in their catalog + policies (knowledge), connected to their store (MCP tools), with memory of the conversation.
- Support + sales + admin help, 24/7, billed by actual usage.
Positioning (one line)
"An AI agent for every store — grounded in your catalog, connected to your CMS, billed by usage, not by always-on bots."
This is the why behind every architectural decision in these docs. When in doubt, optimize for: make it trivial and cheap to give a capable agent to one more store.