HardcoreAgents.io

Problem

I kept writing the same infrastructure code across every client project. Auth, orchestration, deployment, database, monitoring, backups — each build reinvented the plumbing. Each client got a bespoke snowflake that was hard to maintain and impossible to reuse.

Worse: every client needed specialized AI agents that were 80% the same under the hood. Every project needed a frontend-dev agent, a backend-dev agent, a devops agent, a CRM-ops agent. Building them per-client was waste.

The ask I gave myself: build the orchestration layer once, deploy it as a multi-tenant platform, and let every future project (mine and my clients') compose the same agent library.

Approach

HardcoreAgents.io is a production multi-tenant platform. One Dokploy deployment on a Hostinger VPS runs the FastAPI orchestration backend, a Next.js 16 dashboard with a Monaco code editor, a PostgreSQL primary with Redis cache, and a fleet of 14 specialized agents addressable over a consistent API.

Every agent is a first-class citizen with a defined responsibility, trigger rules, and MCP integrations. The platform handles auth (Firebase), orchestration (FastAPI + OpenAI Agents SDK), cost tracking, slow-query diagnostics, cron-driven Airtable ↔ Supabase sync, and zero-downtime rolling restarts.

The Agent Fleet

Fourteen deployed agents today:

• hardcore-frontend-dev — React 19 + Vite/Next + Tailwind v4 + shadcn + WCAG
• hardcore-backend-dev — FastAPI + Pydantic v2 + asyncpg + Supabase + OpenAI Agents SDK
• hardcore-devops — Docker + Traefik + Dokploy + VPS automation
• hardcore-crm-ops — Airtable operations + sync pipeline (via Airtable MCP)
• hardcore-content-engine — text/content synthesis with HVC brand voice validation
• hardcore-report-gen — analytics & reporting (Pandas + chart generation)
• hardcore-agent-builder — scaffolds new agents from specs
• hardcore-automation-mapper — maps business workflows to automation solutions
• hardcore-keyword-intel — keyword research and SEO strategy
• bug-hunter — issue triage and regression detection
• doc-synthesizer — document summarization and extraction
• e2e-tester — Playwright-driven end-to-end testing
• knowledge-pipeline — unstructured markdown → structured assets
• structure-researcher — codebase exploration and architecture mapping

Each agent has its own trigger rules, MCP integrations, and output contracts. The platform routes requests to the right agent; the agent handles the specialized work.

Build Notes

Traefik + Dokploy for push-to-deploy. Push to main, Dokploy rebuilds the Next.js standalone artifact and rolls the container. Traefik handles TLS and subdomain routing to both hardcoreagents.io and hardcoreagents.ai from the same host.

asyncpg, not an ORM. Parameterized SQL only. The performance delta vs. SQLAlchemy for the orchestration workload is measurable, and the schema churn is low enough that the ORM cost doesn't pay for itself.

Diagnostics API as a first-class endpoint. /api/v1/diagnostics/* exposes slow query detection, cost trends, and container stats. The portfolio you're reading hits this endpoint on a 5-minute ISR cache to render the live stats rail on the home page.

Monaco in the dashboard. Agents are configured via Monaco code editor inside the dashboard — YAML config, test prompts, output schema, all edited in-browser with syntax highlighting.

Daily pg_dump + 7-day rotation. Nothing fancy. A cron, a dump, a rotation. Recovering from ransomware or a bad migration is a one-liner.

Results

Production multi-tenant platform serving two live domains. Every new client project bootstraps against the agent fleet instead of building from scratch. The portfolio site you're reading is a showcase of the same stack HardcoreAgents.io runs on — Next.js 16, Tailwind v4, FastAPI patterns, Dokploy deployment, Hostinger VPS.