01IN PRODUCTION

Self-Hosted AI Agent Pipeline

On-prem LLM agents automating real estate lead qualification.

AUTOMATIONAIPROXMOX

§ OVERVIEW

n8n-based workflow that automates real estate lead qualification and follow-up using LLM agents running on-prem.

§ STACK

Proxmox VE

Virtualization host — runs all LXC containers

Ollama

Local LLM runtime (Mistral 7B, Llama 3)

n8n

Workflow orchestration and trigger management

Postgres

Lead data persistence and deduplication

Python

Custom scoring scripts invoked by n8n webhooks

Cloudflare

Tunnel for secure inbound webhook delivery

§ DECISIONS

Local LLMs over OpenAI API

Client lead data includes names, addresses, and financial details. Sending that to a third-party API creates a compliance surface I don't want. Local Ollama keeps inference on-prem — zero data egress, no per-token cost at scale.

n8n over LangChain

LangChain is built for engineers. n8n gives me a visual flow I can hand off or explain to a non-technical client without a 30-minute setup walk. Ops-friendly > developer-flexible for this use case.

Proxmox LXC over Docker Compose

LXC containers give me snapshot-level backup with VZDump, resource isolation without container overhead, and live migration if I scale to a second node. Docker Compose would've been faster to set up and harder to recover.

§ WHAT'S NEXT

•Multi-model routing — route complex qualification tasks to larger models, simple tasks to faster ones
•SMS follow-up integration via Twilio webhook into n8n
•Lead scoring dashboard in Metabase pulling from Postgres
•Formalize the pipeline into a white-label product for other agents

Zero-Trust Home Lab →