The bet

Conversational agents are becoming a buying channel. Yet a merchant catalogue is not consumable by an LLM: no tools, no interface contract, no isolation. CONVOY positions itself as the “backend for LLMs” that fills that gap.

My role

Co-founder and CTO: technical architecture, design of the exposure surfaces, first prospects and client onboarding.

Architecture

• Symfony / API Platform for the business domain and multi-tenant administration.
• FastMCP (Python) for the MCP servers exposed to agents.
• Seven exposure surfaces: Web, App, Browser, Shopify, Embed, Page, AI — each with its own contract.
• Generic tool naming conventions, reusable from one client to the next without rewriting integrations.

Traction

First prospects and onboarding under way in the e-commerce and legaltech sectors.

Architecture

Agents never hit the merchant catalogue directly: they go through generically named MCP tools served by FastMCP, on top of a Symfony / API Platform business core isolated per merchant.

Ordering through conversation

Embed surface: the widget grafts onto the existing store. The agent presents the catalogue, the buyer adds to the cart without leaving the conversation, and the conversation room follows them from one assistant to the next.

Native micro-app: when the conversation has to give way to a form — quantities, address, confirmation — the order is validated by the business core, not by the agent.