# Local AI vs. Paid AI — My Real-World Experience

**URL:** <https://community.homey.app/t/local-ai-vs-paid-ai-my-real-world-experience/161641>\
**Category:** Tutorials\
**Created:** [November 1, 2026, 4:29pm UTC](https://community.homey.app/t/local-ai-vs-paid-ai-my-real-world-experience/161641 "2026-11-01T16:29:35Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Percy\_Mckinney](https://sea1.discourse-cdn.com/flex025/user_avatar/community.homey.app/percy_mckinney/32/174103_2.png) [@Percy\_Mckinney](https://community.homey.app/u/Percy_Mckinney)\
**Post date:** [November 1, 2026, 4:29pm UTC](https://community.homey.app/t/local-ai-vs-paid-ai-my-real-world-experience/161641/1 "2026-11-01T16:29:35Z")

</div>

# Local AI vs. Paid AI — My Real-World Experience

There’s been some interest in adding local AI support to LLMVision for Homey Pro. It’s a reasonable request, but I wanted to share my experience and why I currently use paid AI services.

**First, I’m not debating privacy or the ability to operate without an internet connection.** Those are legitimate reasons to choose local AI.

I’m looking strictly at cost, performance, and practicality.

### My experience with local AI

I previously ran Ollama on a local server. When I considered the electricity required to keep that server running, the operating cost exceeded what I was spending on paid AI services.

I also investigated purchasing an Apple device specifically for local AI. Once I considered the hardware investment, it became a very expensive alternative for my needs.

### My current setup

I have **20 security cameras across two properties** , and my AI analysis costs approximately **$5 per month**.

That’s about the price of a latte. ☕

The key isn’t just which AI provider you use. **It’s how efficiently you use it.**

### One event. One AI call.

With LLMVision for Homey Pro, I don’t send a separate AI request every time a camera detects movement.

Instead, I collect multiple frames from the cameras involved and analyze the complete sequence as a single event.

For example:

1. An Amazon delivery vehicle enters the driveway.

2. Two couriers walk to the front door.

3. They deliver packages.

4. They return to the vehicle.

5. The vehicle departs.

**That’s one event and one AI analysis—not five separate AI calls.**

Here’s an actual result from my system:

> **Amazon delivery**
> 
> **Activity:** An Amazon delivery vehicle arrives in the driveway, and two couriers deliver packages to the front door before departing.
> 
> **Best Image Camera:** Front Door Camera
> 
> **Security Level:** Low

The system also selects the best representative image, which I display alongside the written summary in my Home Assistant security dashboard.

### My takeaway

I’m not suggesting that paid AI is better for everyone. If you already have suitable hardware, need local processing, or have much higher API usage, the economics may be different.

But **local AI isn’t automatically the less expensive choice** once you account for hardware, electricity, and operating costs.

For my installation, paid AI delivers the results I want for approximately $5 a month, without purchasing or maintaining another dedicated AI server.

And that’s why I’m currently satisfied with the paid-provider approach.

**I’d be interested in hearing from others running local vision models: What are your actual hardware, electricity, and operating costs?**

I’m particularly interested in real-world comparisons rather than theoretical benchmarks.

**LLMVision for Homey Pro:** [GitHub - mckinneypl/homey-llmvision: LLMVision for Homey Pro — AI-powered multi-camera analysis and Advanced Flow integration. · GitHub](https://github.com/mckinneypl/homey-llmvision)