AI on your own hardware

Run AI on your own computer, or rent it?

You can run a capable AI model on a desktop in your office today, for free, with nothing leaving the building. Or you can rent one for a monthly seat fee and never think about hardware. This table puts three local tools next to two hosted ones on the questions that actually decide it: what computer it needs, where the data goes, and who keeps it patched.

Checked against each vendor's own pages on October 14, 2026. Every cell links to its source. Prices and plans change often, so click through before you buy.

Ollama

Ollama

What computer it needs
No minimum RAM or VRAM published. GPU needs NVIDIA driver 550+ or Apple Metal.Docs also say NVIDIA compute capability 5.0+, and 5.0 through 6.2 cards need driver 570 or newer. AMD needs ROCm v7. The FAQ gives no minimum specs and says models must fit fully in VRAM for concurrent GPU loads. No minimum OS version is published. Cloud models need no local hardware.Ollama GPU docs Ollama FAQ
Does your data leave the building?
Local models stay on your machine; cloud models run on Ollama's servers.FAQ: cloud prompts are processed but not stored, logged or trained on. Ollama collects basic account info and limited usage metadata, not prompt content. No telemetry default is described. Cloud features can be turned off with OLLAMA_NO_CLOUD=1.Ollama FAQ Ollama pricing
Software cost, 2 people
$0 for local use (MIT license). Optional cloud: Pro is $20/mo, or $200/yr billed annually, per account.The Pro price is shown as '$20 / mo. or $200/yr billed annually'. The page also shows $16.67/mo billed annually. The pricing page says Ollama is one account per person, so 2 people on Pro would be $40/mo monthly, or $400/yr billed annually (math is an inference). Free plan is $0 with starter usage credits and no dollar amount published.Ollama pricing
Software cost, 20 people
$0 for local use. Team cloud plan is $500/mo for unlimited users.Team is labeled Early access, includes $1,000 of usage credits per month shared across the team, and lists Unlimited users. For comparison, 20 separate Pro accounts at $20/mo would be $400/mo (inference from one account per person).Ollama pricing
Who keeps it patched
Mac and Windows apps update automatically; Linux users re-run the install script.On Mac and Windows you click the taskbar or menubar item and choose 'Restart to update'. The FAQ does not describe how model files are updated.Ollama FAQ
License
MIT license (open source), per the GitHub repo.No business-use limits appear on the repo page. Cloud plan terms were not checked.Ollama GitHub repo
Our pick
The pick for the office box: MIT license, automatic updates on Mac and Windows, nothing leaves unless you choose a cloud model.

LM Studio

Element Labs

What computer it needs
Mac: Apple silicon (M1 to M4), macOS 14.0 or newer required, 16GB+ RAM recommended. Windows: x64 (AVX2 required) or ARM, 16GB RAM and 4GB dedicated GPU VRAM recommended. Linux: Ubuntu 20.04 or newer.Intel Macs are not supported. Linux x64 or ARM64 AppImage.LM Studio system requirements
Does your data leave the building?
Local use keeps prompts and documents on your device.Privacy policy says no telemetry; update checks send app version, build and OS, and the CDN sees your IP address. Cloud processing (Bionic, Secure Cloud) happens only if you choose it; LM Link to remote devices is not described on the Bionic page.LM Studio privacy policy LM Studio Bionic docs
Software cost, 2 people
$0 on the Free plan (runs local models). Bionic+ is $20 per month per account; Pro is $100 per month.Prices listed per month; yearly billing not mentioned. Two Bionic+ accounts would be $40 per month. Page does not price the desktop app separately; Free plan includes local LLMs. Team subscription plans are listed as coming soon.LM Studio pricing
Software cost, 20 people
No team subscription price published. 20 Bionic+ accounts would be $400 per month.$20 x 20 = $400, individual accounts, not a team rate. Page offers centralized billing for teams using inference credits and says subscription plans for teams are coming soon.LM Studio pricing
Who keeps it patched
Built-in updater on macOS and Windows checks at startup and prompts you to install now or later. Linux has no in-app updater yet.Needs internet for the updater. Docs say the Linux in-app updater is in the works.LM Studio offline operation docs
License
Commercial terms, no license name. Personal and internal business use only.Terms grant a non-exclusive, non-transferable license for personal and / or internal business purposes. They bar sublicensing, distribution, sale, service bureau use, and modifying or creating derivative works. No open source license is named.LM Studio terms
Our pick
The friendliest local app, with 16 GB RAM recommended. Business use is allowed under its terms.

Open WebUI

Open WebUI

What computer it needs
No CPU, GPU or OS minimum published. Docs call 1 GB RAM very low for RAG or embeddings, and list a low-end target of Raspberry Pi / under 4 GB RAMModel memory is extra and depends on the model you run. The '2 GB or more' wording was not seen on the page, so it is not repeated.Performance and RAM
Does your data leave the building?
Runs on your hardware; cloud models get data only if you add one. Telemetry default not stated on page we checkedHomepage says data stays where it belongs and mentions air-gapped deployment. Homepage does not cover telemetry. Cloud model providers you connect would receive prompts.Open WebUI homepage
Software cost, 2 people
$0 per month. Free under the Open WebUI LicenseBranding must stay in place for deployments above 50 users in a 30-day period. Enterprise page says organizational use goes through enterprise licensing; no price published. Hardware excluded.Open WebUI License
Software cost, 20 people
$0 per month published. No per-seat fee listed; free use requires original branding intactEnterprise license (custom branding, support) has no published price; contact sales. Enterprise page also says organizational engagement occurs through enterprise licensing, so a company may be asked to buy one. The Computer product page was not verified and is dropped.Open WebUI Enterprise Open WebUI License
Who keeps it patched
You update it yourself: pull a new Docker image, run docker compose pull, or run pip install -U open-webuiWatchtower can automate pulls, but the docs warn automated updates can break a deployment.Updating Open WebUI
License
Open WebUI License: free, but Open WebUI name and logo must stay in place for deployments above 50 usersBranding rule applies to 50+ users in aggregate. Custom branding or white-labeling needs a proprietary or enterprise license.Open WebUI License Open WebUI Enterprise
Our pick
The shared chat front end for a team on one Ollama box. Someone has to pull the Docker updates.

ChatGPT Business

OpenAI

What computer it needs
Not publishedPage lists desktop and mobile apps but gives no RAM, GPU or OS requirement.ChatGPT pricing
Does your data leave the building?
Yes. Cloud chats run on OpenAI's remote machines, not on your computer.Inferred from the page's wording on cloud virtual machines; the page does not state a prompt-processing policy in so many words. Page also says 'No training on your business data by default.' No telemetry default found.ChatGPT pricing
Software cost, 2 people
$40/mo billed annually ($20/user); $50/mo billed monthlyMath: 2 x $20 = $40 annual; 2 x $25 = $50 monthly. Page states 2+ users minimum. Hardware excluded. The headline $20 is the annual-billing rate.ChatGPT pricing
Software cost, 20 people
$400/mo billed annually ($20/user); $500/mo billed monthlyMath: 20 x $20 = $400 annual; 20 x $25 = $500 monthly. The $25 monthly rate appears only in the footnote.ChatGPT pricing
Who keeps it patched
Not publishedPage does not describe update delivery.ChatGPT pricing
License
Commercial per-seat subscription: $20/user/month billed annually or $25/user/month billed monthly, 2+ users. Full terms of use not published on the page checked.Page states no training on business data by default. No business-use restrictions are listed on this page.ChatGPT pricing
Our pick
$20 a seat billed annually, no hardware, no training on your data. The rental option when nobody owns the box.

Microsoft 365 Copilot

Microsoft

What computer it needs
No RAM or CPU minimum published. Needs a current browser or Microsoft 365 Apps.The requirements page lists browsers, cookies, identity and network needs, but no processor, RAM or disk minimums for Copilot.Microsoft Copilot requirements
Does your data leave the building?
Yes. Prompts and document content are sent to Microsoft's cloud for processing.Microsoft says prompts, responses and Graph data are not used to train its foundation models, and it stores interaction history that admins can view through Purview. Web search sends a generated query to Bing when allowed. Processing can occur in other regions during high utilization.Data, Privacy, and Security for Microsoft Cop...
Software cost, 2 people
$36 per month for 2 users, billed yearly at $18 each, plus a Business plan.The $18 price is a first-year promotion for offers running July 1 to December 31, 2026, and the add-on needs an eligible Microsoft 365 Business plan; monthly billing is $25.20 per user.Microsoft 365 Copilot pricing
Software cost, 20 people
$360 per month for 20 users, billed yearly at $18 each, plus Business plans.Math: 20 x $18.00 = $360 per month, paid yearly, first year only; monthly billing would be 20 x $25.20 = $504 per month.Microsoft 365 Copilot pricing
Who keeps it patched
Microsoft sends updates monthly. Admins pick the channel (default Current Channel).The AI models behind Copilot are updated by Microsoft as well; the privacy page says they are regularly updated and enhanced.Overview of update channels for Microsoft 365... Data, Privacy, and Security for Microsoft Cop...
License
Commercial subscription. Annual commitment, billed monthly or yearly.Business add-on: capped at 300 seats per tenant, needs an eligible Business plan, and you cannot move to an Enterprise plan until the commitment ends. Usage restrictions for businesses were not found on the pages we checked.Microsoft Copilot Business FAQ
Our pick
$21 list, $18 promo through December, on top of 365. Updates handled, residency available. For a team already in Teams and Outlook.

Local AI is a choice about who does the IT

The software is free. The cost is a computer with enough memory and a person who keeps it updated. Rented AI is the opposite trade.

  1. 01Go local when the data cannot leaveClient files under a confidentiality duty, health records, legal matters. A local model answers the question without creating a disclosure.
  2. 02Rent when nobody owns the boxIf no one on the team will update software or restart a server, a hosted seat is cheaper than the first outage.
  3. 03Count the seatsAt two people, the hosted plans cost less than a lunch. At twenty, they run $360 to $400 a month, so a dedicated machine can pay for itself within a year, if someone runs it.

The local tools here can also talk to hosted models when you want them to. The useful setup is often both: local for sensitive work, hosted for everything else.

Spearworks AI

Want a private AI box in the office?

I set up Ollama and Open WebUI on a machine you own, load the right models for the work, and leave a one-page runbook so you or your IT provider can keep it patched.

Book a free consult

Nathan Spear · spearworksai.com

Spearworks Labs · All comparisons · Checked October 14, 2026. Not affiliated with any vendor listed.