Local AI Options, Compared for Non-Technical Buyers
Inclusion criteria, stated up front: an option appears here if it (a) runs models on hardware you own, (b) is genuinely obtainable today, and (c) is plausible for a small office rather than a research lab. Categories are not interchangeable — a runtime is not an interface is not a hardware tier — so they are compared within category, never against each other. Entries marked hands-on: no are curated from documentation and community evidence, not our own bench testing yet; we say which is which.
Software: runtimes and interfaces
| Option | What it is | Data exposure | Setup complexity | Fits best | Hands-on |
|---|---|---|---|---|---|
| Ollama | The de-facto standard runtime for open models | Local by default; optional cloud features exist — a setup choice to make deliberately | Low–moderate: installable in minutes, tuning takes longer | Almost every local setup; the default starting point | yes |
| Open WebUI | Chat interface your staff use in a browser | Local — served from your own machine | Moderate: pairs with a runtime; office-network setup is where IT help earns its keep | Teams — gives everyone the familiar chat experience | yes |
| LM Studio | Desktop app: runtime + interface in one | Local | Low: closest to "install an app" | A single user evaluating on their own computer | no |
Hardware tiers
| Tier | What it looks like | What it runs | Fits best |
|---|---|---|---|
| Existing modern computer | A recent machine with 16 GB+ memory | Small models — real but modest capability | Trying local AI before spending anything |
| Workstation-class + GPU | A strong desktop with a consumer graphics card | Mid-size models — the daily-driver tier for drafting and document Q&A | A small office's first real deployment |
| Dedicated small server | A quiet box serving the whole team | Larger models, multiple simultaneous users | Teams past ~10 regular users |
Model families themselves (what to actually run) change too fast for a static table to serve you honestly — that guidance lives in the guides, where each recommendation carries its verification date.
How to use this page
Pick your hardware tier from your team size, pick the software pair that matches your IT support level, then take the combination into the cost worksheet. If the numbers hold up, the assessment turns the combination into a concrete plan.
Next step
Wondering if this fits your office?
The readiness assessment walks through your data sensitivity, current AI use, and what a local setup would actually involve — with an engineer, not a salesperson.
Assess your readiness →Frequently Asked Questions
Ask about this article
Get a plain-language answer drawn from this article. Answers are AI-generated from the text on this page.
Related Templates
External Resources
Authoritative references and tools related to this documentation type.