It was not a reply bot. It was a digital employee per business.
EtherCode OS runs on Cloudflare Workers and operates per demoId to provide business-specific context, conversational memory, and commerce execution connected to catalog and checkout.
Real problem#
Commercial teams lose opportunities when every conversation depends on human operators without continuous context, useful memory, or direct catalog/checkout integration.
Solution#
I designed a multi-tenant architecture by demoId combining intent classification, lead memory, dynamic context, and response guardrails to convert conversations into qualified opportunities.
What the app is#
It is an AI backend for commercial assistance and lead capture built on Cloudflare Workers. It acts as a Digital Employee per business demo (demoId), with conversational memory, response rules, and ecommerce/payment support.
Technical base (stack)#
Runtime: Cloudflare Workers + Durable Objects + D1 + KV.
AI: Cloudflare AI binding (configurable model, default @cf/meta/llama-3-8b-instruct).
Integrations: Tienda Nube, Mercado Pago, Supabase.
Main code: src/index.js and wrangler.toml.
How it works (technical flow)#
Client calls /chat or /v1/demo/chat.
Rate limiting is applied per IP (RateLimiterDO) and per session/day (ChatSessionDO).
In demo mode, business profile and dynamic context are loaded (agent-context-service.js).
Commercial enrichment with Tienda Nube catalog and Mercado Pago checkout (commerce-runtime-service.js).
Intent is classified, structured context is built, and LLM is invoked.
Response is validated/sanitized (length, forbidden phrases, CTA rules, and no price hallucinations).
Messages are persisted in D1 and returned through SSE streaming.
Integrated modules#
Conversation engine and API router: src/index.js.
Intent classification: intent-classifier.js.
Lead operational memory: memory-summary.js.
LLM context builder: context-builder.js.
Prompt/policy/validation: system-prompt.js, response-policy.js, response-validator.js.
External integrations: tiendanube-client.js, mercadopago-client.js, supabase-source.js.
Database schema: migrations.
Functional endpoints#
Health: /health, /v1/health.
General chat: /chat, /v1/chat.
Per-business demo chat: /v1/demo/chat.
Lead capture: /lead, /v1/lead.
Business config: /v1/demo/business, /v1/demo/business/get, /v1/demo/business/list.
Runtime/admin config: /v1/admin/agent/source, /v1/admin/agent/integration, /v1/admin/messages, /v1/admin/runtime/ip-rate-limit.
Public demo view: /v1/demo/public.
Architecture#
Serverless system with per-business isolation (demoId), state control via Durable Objects, lightweight persistence with D1/KV, and real-time commerce integration that turns conversation into action.
Runtime
Cloudflare Workers + Durable Objects
Data
D1 for conversation, KV for fast state
AI
Cloudflare AI binding with configurable model
Commerce
Tienda Nube + Mercado Pago + Supabase
Strengths
Scalable serverless architecture with low operational cost.
Per-demoId multi-tenancy with business context isolation.
Useful conversational memory for lead qualification.
Real commerce integration: catalog + checkout.
Response guardrails for commercial quality and risk control.
Operational governance through admin endpoints protected by X-Admin-Key.
Result#
EtherCode OS enables deploying a digital employee per business with dedicated context and operational control, reducing commercial friction and accelerating opportunity capture.
| Capability | Before | With EtherCode OS |
|---|---|---|
| Per-business context | Manual and fragmented | Automatic by demoId |
| Commercial support | Operator-dependent | Agent-assisted with guardrails |
| Sales integration | Disconnected | Catalog and checkout connected |
Project summary#
I built EtherCode OS as a B2B conversational backend on Cloudflare Workers, with per-demoId multi-tenant architecture, conversational memory, and response guardrails. I integrated real-time catalog (Tienda Nube), checkout (Mercado Pago), and external context (Supabase), plus admin endpoints for runtime governance and observability.
If you are hiring to build production AI platforms, I want to contribute to that team.
I focus on robust architecture, technical execution, and measurable outcomes in real systems.