The Local AI Stack Audit: Everything You Need, What It Costs

Stack audit editorial diagram: five layers, $0 monthly bill

News · August 22, 2026 · Operator Co

We run our entire business on a local AI stack: search, scraping, vector storage, automation, image generation, and the language models themselves. No per-token bills, no vendor lock-in, no data leaving the machine. People ask what it actually takes, so here is the honest audit: every component, what it does, and the real numbers.

The components

Ollama

Runs the language models. One command to pull a model, one to serve it. The models do everything from drafting to code review on a single GPU.

SearXNG

Private meta-search. Aggregates results from dozens of engines without tracking, and returns clean JSON your automations can consume.

Crawl4AI

Turns any web page into clean markdown for your pipelines. JavaScript-heavy sites included, via a bundled browser.

Qdrant

Vector database. Stores embeddings of everything you research so your models can cite their own sources instead of guessing.

n8n

Workflow automation, self-hosted. Connects the whole stack: watch a feed, crawl it, summarize it, ship it, all without touching a cloud console.

ComfyUI

Image generation on your own GPU. Product covers, social visuals, textures. Deterministic workflows you can rerun and version like code.

The cost math

The hardware is one used RTX 3090 (24GB) at roughly €800. Amortized over three years that is about €22 a month, and electricity adds a few euros depending on use. Everything else on this list is free and open source. Compare that with typical cloud AI subscriptions at €100–€200 a month, which is €3,600–€7,200 over the same three years, and the break-even lands in month four to six.

The second number people miss is data. With local inference, customer documents and research never leave your machine. For EU-based work that is not a nice-to-have; GDPR makes it a selling point you can put in a proposal.

What the audit checks

Our free Local AI Stack Audit walks through your current toolset and answers three questions: which cloud subscriptions map to a local replacement, what hardware you actually need for your workload, and where the migration risks hide. It is the same checklist we used to move our own operations off the cloud, condensed into a scorecard you can act on.

Get the free audit →