Case study

Agents API Hub

Public Vercel demo interface of the API Hub managed through Hermes.

Connection API between my AI agents (Hermes) and my products: a single authenticated, documented entry point exposing targeted reads and controlled actions, with audit and human approval.

Context

Central service connecting Hermes to several of my products: e-commerce, an affiliate site, this portfolio and a voice app. Each project keeps its own database and business logic; the hub only handles the connection.

Problem

Give AI agents useful access to projects without opening up their databases, without multiplying connections, and while keeping a record of every action.

Solution

A single API that exposes agent-safe capabilities, domain by domain: targeted reads, controlled actions, events, audit and human approval. It routes to each project’s endpoints and publishes a single OpenAPI documentation.

My role

Architecture, API development, access security, PostgreSQL database, product integration and deployment.

Architecture

Hono with typed routes and OpenAPI documentation generated from Zod
PostgreSQL and Drizzle for API keys, audit and events
Signed (HMAC) connections to each project’s backend
Docker Compose, Nginx and Certbot on Hetzner

Challenges addressed

Never turning the hub into a central database
Precisely limiting what each key can do
Logging every agent action
Keeping secrets on the server, including for Majesska’s voice

Verifiable results

Single entry point for four domains: e-commerce, affiliate, portfolio and voice
Automatically generated OpenAPI documentation
Unified event queue with acknowledgment
Voice gateway (speech synthesis, transcription, tutor replies) used by Majesska

Next steps

Keep documenting API contracts and measuring workflow reliability without publishing sensitive operational data.