LM Studio Repriced My AI Stack to Zero

Show Notes

Watch the video: https://www.youtube.com/watch?v=0HNpaQ0vAXc

LM Studio turns any laptop into a free OpenAI-compatible endpoint, and 80% of what your team runs on GPT-4o has no business being on a paid API.
Playbooks + my answers, free - Church of SEO Jesus: https://www.skool.com/church-of-seo-jesus

What you get in this one:
- Why the same eight prompts eat most of your OpenAI invoice
- The exact base-URL swap that repoints Cursor, Zapier, and Python scripts at localhost
- The split that keeps working: what moves local, what stays on Claude/GPT
- The 50M-token math ($625 vs $7.50) your finance person will want to see
- The $500 used Mac Mini setup that answers every summary request forever

Open LM Studio yourself, click the Developer tab, hit Start Server, and change the base URL in any tool you already pay for to http://localhost:1234/v1. Paste any string as the API key. That is the proof, not me telling you.

Sources:
- LM Studio: https://lmstudio.ai
- OpenAI pricing: https://openai.com/api/pricing

Chapters:
00:00 LM Studio killed the AI line on my invoice
00:28 Why nobody on my team noticed the switch
01:00 Who this is for (operators signing AI bills)
01:15 The same eight prompts eating your OpenAI account
02:00 The 80% cut is real
02:12 Subscribe + Free Church ask
02:37 The LM Studio Developer tab and localhost:1234/v1
03:08 Verify it yourself in Cursor right now
03:25 What moves local: summaries, rewrites, tags, drafts
04:02 What stays on Claude and GPT-4o
04:33 GPT-4o-mini vs GPT-4o at 50M tokens: $7.50 vs $625
05:15 The under-$100 / $1k / $1k+ audit rule
05:37 The $500 used M2 Mac Mini under the desk
06:08 lms CLI, cheap VPS, gaming laptop paths
06:33 The 80% number and why it exists
06:50 The six-month window before competitors catch up
07:08 Next: the open-model shortlist by workload

Audio edition of the Stewart Vickers AI YouTube video of the same name. Full transcript on this episode page.

What is Stewart Vickers AI?

The audio edition of the Stewart Vickers AI YouTube channel. Each episode takes one AI release - a model, a tool, a price cut - and translates it into money for business owners: what got cheaper, who got undercut, and what to do about it this week. Operator credibility over theory, specific numbers over vibes. Every episode links the original video and the free Church of SEO Jesus community where the prompts and workflows live.

Look at the AI line on my invoice from last month. It stopped growing. Not because we stopped using AI, because I moved 80% of it off OpenAI and Anthropic and onto a free AIPI that was already sitting on the laptop I own. The tool is LM Studio. It runs a local, OpenAI-compatible endpoint on your own machine. Every tool that accepts a base URL, cursor, Zapier, an internal chatbot, a Python script, points at it in one line.

None of them notice. I'm Stuart, founder-operator. I've been running this switch for a month across four different tools, and my team has not once asked what changed, because nothing they can see changed. The output feels the same, the latency is often better because there's no network call, and the bill for those workloads went to zero. Fair warning up front. Most of the local Modal coverage online is aimed at hobbyists running chatbots on gaming PCs, or at IT admins worried about privacy.

My video is for the person actually signing the AI invoices every month, and quietly wondering how it climbed to four figures without anyone deciding it should. Here is what most operators miss. Your team is not running your OpenAI account at full strength. Nowhere close. They are running the same eight prompts on it, over and over. Summarize a customer email, rewrite it in a friendlier tone.

Tag inbound tickets by category. Draft the reply to a vendor. Four down. Extract three bullet points from a call transcript. Classify a fresh lead. Clean up meeting notes.

Pull the action items out of the dock. Eight total. Same eight every week. None of that needs frontier reasoning. It needs a Modal that can hold a sentence together and follow instructions. An eight billion parameter local Modal does that fine.

It has been doing it fine for about a year now, and it costs nothing PR call because it runs on hardware you already paid for. So the answer, the one the title promises, is yes. You can cut your AI bill by roughly 80% this week, without asking your developers to rebuild anything, without your team noticing the change. One quick ask before the how. Subscribe on this channel, and every new open Modal or local inference route lands on your feed the day it ships, instead of your competitors. And the exact commands and config snippets I'm about to show sit in the pinned comment under this video.

The full playbook version lives in the church of SEO Jesus, my free school community. Free to join. Link is below. The switch itself. LM Studio ships a developer tab. You hit start server.

It exposes an OpenAI compatible endpoint at localhost, port 1234. In cursor, you open settings, paste that URL as the OpenAI base URL, and type any string as the AYPI key. Any string. It doesn't check it. Cursor accepts it. It has no idea it is talking to a Modal on your own desk.

The endpoint speaks the same JSON as OpenAI's. Every tool that speaks OpenAI now speaks to your laptop. You do not have to trust any of this. Open LM Studio yourself right now, click the developer tab, hit start server, and change the base URL in whatever tool you already pay for to localhost colon 1234 slash v1. Proof not me telling you. Now the split.

What moves local? What stays paid? Boring, high volume, low stakes calls move local. Summaries, rewrites, tags, classifications, first draft emails, internal search over your own docs, meeting note cleanup. Anything a junior on your team could do without escalating. Anything you'd hand to a summer intern with a checklist and a slack channel to ask questions in.

Your 80% hides in there, and it does not care whether it is a frontier Modal or an 8 billion parameter one running on your desk. Now what stays paid? Frontier reasoning. Code that ships to production, legal or medical work where a mistake costs real money. Long context analysis when you genuinely need 200,000 tokens of memory in one call. Anything a customer will see before you review it.

Your 20% lives here, and it stays on Claude or GPG because the quality gap between an 8 billion local Modal and a thousand billion frontier Modal is real, and it is exactly where it should be. The math your finance person cares about. 50. Sounds cheap? Nobody notices. 50 million input tokens on GPG-4o the same volume is roughly $625, and that is one team.

One month, multiply by 12, multiply by every team in your told them to switch down to mini. My rule of thumb. If your total monthly AI spend is under a hundred bucks, do not bother, you have bigger levers. If it is between a hundred and a thousand, do the audit and see. If it is over a thousand a month, half of that number is being paid to summarize things that a used Mac mini could summarize for free, and you are the last person in the loop to notice. About that Mac mini, a used M2, 16 gigs of unified memory, about 500 quid on eBay.

It runs an 8 billion parameter Modal at 20 to 40 tokens PR second, fast enough that nobody on your team will notice a delay. You leave it plugged in under a desk somewhere. It answers every summary request your business fires at it forever for the price of the electricity. If you don't want to buy hardware, LM studio also ships a CLI called LMS and a Python SDK. So a cheap VPS runs the same setup for the price of a couple of coffees a month. If you're on windows or Linux, LM studio ships the same developer tab and the same one click server on both.

A four-year-old gaming laptop with 16 gigs of RAM runs the same Modal, slower than the M2, still faster than a 80% is the share of your AI invoice that has no business being on your AI invoice. It is being paid because nobody sat down and split the work into what needs a genius and what needs a competent intern. The window for moving it is right now, while your competitors are still assuming local models are a hobbyist toy, and while their agencies are still quietly marking up the open AI pass-through on their monthly retainer. Six months from now, everyone in your inbox signing an invoice will have already done this. If you want the exact list of which open Modal matches which workload and where each one starts to break, I ran the tests, I graded them and I have the shortlist in the next video. Watch that one now before your next invoice lands.