// REFERENCE PROJECT: E-COMMERCE · AGENTIC AI PLATFORM

Vendra

One AI platform meant to replace a scattered e-commerce tool stack.

E-commerce founders often run a support chatbot, an ads manager, a CRM and more subscriptions that do not talk to each other. Vendra is meant to bring these jobs into one AI platform. We built the foundation first: an AI sales and support chat with product-catalogue context, RAG answers and quality review, plus a control layer with an MCP-compatible endpoint that further agents are meant to plug into. Further agents are planned.

Vendra

AI Commerce OS

Sales chat

Social DMs

Google Ads

Meta Ads

SEO / GEO

CRM

Site builder

Agent-ControlMCP

Architecture sketch: the tiles show the planned channels. What is built so far is the foundation of chat and control layer.

Industry

E-commerce · SaaS

Type

AI platform with an agent control layer

Cycle

Foundation built, further agents planned

Stack

Next.js · TypeScript · SQLite/Postgres · RAG · LLM · MCP

// THE STARTING POINT

E-commerce founders lose time wiring together tools that do not cooperate: a chatbot, an ads manager, a CRM, an email tool and an SEO plugin. Vendra’s goal is one platform where agents take over these jobs and a central control layer coordinates them. We started with the piece that helps most day to day, the sales and support chat. The architecture is set up so that every further agent plugs into the same control layer instead of becoming yet another isolated tool.

// WHAT WE BUILT

✓AI sales and support chat: website widget, knowledge base and FAQ, product-catalogue context and RAG answers

✓Admin with conversation review and quality checks

✓Control layer with an MCP-compatible endpoint that further agents are meant to plug into

// HOW THE PROJECT RAN

Phase 1: Goal and scope

Control layer first

We mapped a founder’s real tool stack and what each job actually needs. That set the architecture: first a control layer with an MCP endpoint that chat, ads, SEO and CRM agents are meant to plug into later.

Phase 2: First agent

The sales and support chat

As the first agent we built the chat end to end: website widget, product-catalogue context, RAG answers and an admin with conversation review and quality checks. It was the first agent running through the control layer.

Outlook

Planned: further agents on the same control layer

Agents for social chats, email support, Google and Meta ads, SEO and GEO, CRM, email marketing and a website builder are planned. They are meant to plug into the same control layer.

Outcome

agent built end to end: sales and support chat with admin and quality review

MCP

compatible endpoint of the control layer that further agents are meant to plug into

Planned architecture

The control layer is designed for 8 channels: chat, social, email, Google and Meta ads, SEO, CRM and marketing. Of these, the chat is built; the other channels are planned.

Status and limits

Built and documented is the foundation: the sales and support chat with admin and quality review, and the control layer. Further agents are planned; no rollout is documented on this page. There is no public live address for Vendra here yet.

NEXT PROJECT

GymOrdo →

Want AI agents inside an existing workflow without every tool becoming its own island? Describe your plan and get a first assessment of scope and timeline.