The reference guide · 16 chapters · 87 pages · templates included · free updates for life
Local AI
at work
The AI at Work Kit: THE reference guide for scoping, deploying and running a private local AI server for a team — privacy law and the EU AI Act made workable, multi-user architecture, security hardening, an acceptable-use policy, a structured pilot, a real TCO. For an IT lead, an engineering manager, a founder or a consultant who has to decide, and get others to decide.
Get the kit — $24$24 · one-time payment · free updates for life
What is The AI at Work Kit?
The AI at Work Kit is a 16-chapter reference guide and 18-file template pack for scoping and deploying a private local AI server for a team. It's built for IT leads, engineering managers, founders, and consultants who need to size a Linux server with a shared 24-48GB GPU for 15-20 users. You get the privacy-law and EU AI Act groundwork, an acceptable-use policy, a 30-day pilot plan, and a TCO calculator, plus lifetime updates in your members area.
- 16 chapters and 87 pages, from privacy law and the EU AI Act to a 30-day pilot.
- 18-file template pack: a leadership memo, an acceptable-use policy, and a TCO calculator.
- Anchor profile: 15-20 users on a Linux server with a 24-48GB professional GPU.
- Chapter 11 runs a structured 30-day pilot with success criteria set before day one.
- Chapter 12 gives you a 3-year TCO calculator to compare against cloud AI costs.
- $24 one-time payment, updates for life included, 30-day money-back guarantee.
✓ This is for you if…
you're an IT lead, an engineering manager, a founder or a consultant, your team is already using generative AI (officially or as shadow AI), and you need to scope a deployment for a group — a company that has, or can budget, a server with a GPU, for an internal use case (chat, documents, drafting).
✗ This isn't for you if…
you want to set up an assistant for yourself alone at home (that's the Local AI Kit) or get an agent coding inside your editor (that's the Local Copilot Kit) — and this guide never substitutes for the advice of your own counsel, DPO or security lead on your specific situation.
Why you can trust this
This guide is written and maintained by BestLLMfor.com — tracking 239 locally-runnable models and 73 guides, published on an open methodology. Every version, Ollama tag, license and command in this guide was checked against official sources (ollama.com, Hugging Face model cards, project changelogs and documentation) at the edition date, September 2026. Nothing is presented as measured on our own machines unless the text says so explicitly; performance figures point to the living pages of bestllmfor.com. This edition's US GPU, Mac Studio and mini-PC prices, and its EU/US regulatory references, were checked the same way against EUR-Lex, EIA.gov and manufacturer spec pages, and are re-checked at every edition. What you read is condensed and verified — not a random compilation of tutorials found online. Who's behind BestLLMfor →
The problem
Your team is already using generative AI. Officially or not, with or without your sign-off. The real question isn't "should we allow it?" anymore — it's "who decides where the data goes, and who can prove it?" A local AI server answers both at once — but between privacy law, the EU AI Act, multi-user architecture and an acceptable-use policy nobody has signed yet, the project stays wide open until it's scoped.
The groundwork already exists, scattered across free guides and the team-deployment chapters buried in other guides. This guide is THE reference method: the full path assembled end to end (scoping → architecture → security → adoption), with the templates ready to adapt that no single free guide gives you assembled — not a pile of tutorials you have to cross-check yourself.
Local AI for a team vs per-seat cloud AI
Same everyday use — chat, drafting, questions on internal documents — compared honestly for a 15-20 person team:
| Local AI for a team | Cloud AI (per seat/month) | |
|---|---|---|
| Conversations stay on the company's own server | Yes | No |
| Cost that climbs with headcount | No (near-zero marginal cost) | Yes, per seat |
| Data sent to a third-party processor | None | Depends on the vendor |
| Works offline / with the network segmented | Yes | No |
| Real-time web search | No | Yes |
| Newest, heaviest models, zero maintenance | Limited by the company's own hardware | Yes |
| Data-processing responsibility (privacy law) | Stays with you — local doesn't remove it | Stays with you |
Local removes the third-party processor and cross-border transfer question, not the underlying data-processing responsibility or the need for a written policy — Ch. 2 lays out exactly what changes, and what never does.
What you get
- Your members area, for life — all 16 chapters, readable on the web, always the current edition.
- The full guide as a PDF (87 pages) — the same content, yours forever, offline.
- A 18-file template pack — a leadership memo ready for a decision meeting, an AI acceptable-use policy, an executable privacy checklist, a 30-day pilot plan, a TCO calculator (method + spreadsheet), a hardened server Docker compose file (SSO commented in), and more.
- One clear anchor profile — 15-20 users, a Linux server, a 24-48 GB professional GPU: the sizing numbers never contradict each other from one chapter to the next.
- The "server troubleshooting" chapter — a decision tree by symptom, built for an outage that hits 15-20 people at once, not a single workstation.
- Living edition — every update (law AND tools) shows up in your members area, for life, at no extra cost.
Tools: Ollama, Open WebUI, Docker, SSO/OIDC (Microsoft Entra ID walked through step by step). No line of code required on your side, no hardware price fixed in the PDF (see bestllmfor.com/ai-hardware/), no individualized legal advice.
The 16 chapters of the kit
87 pages, from scoping the case for leadership to a server running in production for your team. Nothing to skip: every chapter links to the next.
- Why Local AI at Work, Now
- Privacy Law and Local AI: GDPR, CCPA/CPRA, HIPAA
- The EU AI Act (and What US Teams Need to Know)
- Multi-User Architecture: The Big Picture
- Ollama + Open WebUI Server: Step-by-Step Deployment
- Authentication and Accounts: SSO, Roles, Lifecycle
- Logging, Backups, Continuity
- Choosing Your Server Hardware
- Server Security and Hardening
- Acceptable-Use Policy for AI at Work
- The 30-Day Pilot: Method and Governance
- TCO vs. Cloud: The Method and the Calculator
- Training Your Team
- Measuring the Value
- Troubleshooting and Day-to-Day Operations
- This Edition Has a Shelf Life — Now What?
Living edition: every update enriches this table of contents (dated update log, ch. 16) — you never repurchase to get the next version, even when the EU AI Act timeline shifts.
Your team's server, step by step
No line of code to write yourself: the server commands are given in full inside the guide, ready to paste. A preview of the path (anchor profile: Ubuntu Server, 24-48 GB professional GPU):
nvidia-smi curl -fsSL https://get.docker.com | sh curl -fsSL https://ollama.com/install.sh | sh sudo systemctl edit ollama # [Service] # Environment="OLLAMA_HOST=0.0.0.0:11434" # Environment="OLLAMA_NUM_PARALLEL=4" sudo systemctl daemon-reload && sudo systemctl restart ollama ollama pull qwen3:32b docker compose -f docker-compose.server.yml up -d
1 · Server and model
GPU checked, Ollama exposed properly, Open WebUI running in Docker with the compose file included — a model answers (Ch. 4-5).
2 · SSO and accounts
Open WebUI wired to your company IdP (Microsoft Entra ID walked through step by step), simple roles, offboarding in one action (Ch. 6).
3 · Security and policy
Reverse proxy plus TLS in front of the API, an acceptable-use policy every user signs before their first login (Ch. 9-10).
4 · Pilot and review
A bounded 30 days, success criteria written before day one, a go/no-go meeting with an honest verdict (Ch. 11).
- Your members area for life: all 16 chapters, always current (full table of contents above)
- The PDF guide (87 pages) + the 18-file template pack (memo, policy, checklists, TCO calculator, Docker compose) to download
- The single anchor profile (15-20 users, 24-48 GB GPU) and the server troubleshooting decision tree
- Every future edition, for life — law AND tools tracked, they show up in your members area automatically
- Instant access after payment, access key emailed to you
All the kits, for life: every guide, current and future, for $49 — already own this one? Upgrade for $25.
Get the kit — $24Stripe receipt emailed automatically; invoice on request by replying to the delivery email.
The honest truth
A local AI server for a team solves privacy exposure and per-seat cost — it doesn't solve everything, and this guide is not legal advice. A reference guide owes you that honest map:
This guide gives no legal advice: Ch. 2 and Ch. 3 give a workable privacy-law and EU AI Act framework, with the official public sources cited, but the final sign-off is always your own counsel's, your DPO's or your security lead's.
By the end, you'll know how to
- Present a clear, honest, numbers-backed memo to your leadership team, and be plain about what this guide covers — and what it doesn't.
- Fill in an executable privacy checklist and clarify your company's status under the EU AI Act (a deployer, not a provider, in the normal case).
- Deploy and walk your team through a multi-user architecture diagram, with an Ollama + Open WebUI server in production, SSO wired in.
- Get an AI acceptable-use policy signed, adapted to your company, before each user's first login.
- Run a 30-day pilot with success criteria written before day one, and price your own 3-year TCO against the cloud.
Living edition — like every BestLLMfor guide
"Free updates for life" isn't a slogan: this guide tracks both the law (EU AI Act, privacy law) and the tools (Ollama, Open WebUI) for you, at no extra cost — the EU AI Act timeline already shifted once between this guide being written and published.
v2026.09 (current edition) — first edition of the guide: 16 chapters, one anchor profile (15-20 users, a Linux server, a 24-48 GB professional GPU), 18-file template pack.
Future editions (new EU AI Act dates, new models, new tools) get added here, dated, and land in your members area at no extra cost.
Frequently asked questions
How is this different from the other guides in the collection?
The Local AI Kit is for one person setting up an assistant on their own machine — none of that is repeated here. The Local Copilot Kit is for a developer coding solo with a local agent. This guide is for the person who has to get a local AI server adopted and kept running for a team: privacy-law and EU AI Act scoping, multi-user server architecture, security, an acceptable-use policy, a pilot and a TCO — ground the other guides don't cover.
Do I need to be a developer, or a systems administrator?
No. No programming skill is required on your side — every server command is given in full, ready to paste. This guide does assume a Linux server with a GPU either already exists or can be budgeted, and that someone technical (you or a teammate) can run the commands from Chapter 5 onward.
Does this guide give me legal advice?
No, and it isn't meant to. Chapters 2 (privacy law) and 3 (the EU AI Act) carry a disclaimer, stated up front and again at the close: these are workable frameworks with the official public sources cited, not individualized legal advice. This guide prepares the case you bring to your own counsel, DPO or security lead — it never substitutes for their sign-off.
What is the difference with the free guides on bestllmfor.com?
The free guides on the site explain each piece on its own — SSO for Open WebUI in one place, hardening an Ollama server in another, with no throughline between them. This guide assembles the full path in order (scoping, architecture, security, adoption), and hands you the asset pack — templates, checklists, a TCO calculator, a Docker compose file — that no single free guide gives you assembled, kept current for life.
Does this guide cover document RAG or coding locally?
No, deliberately. Advanced document RAG (scripted, tuned) belongs to the Local RAG Kit; coding locally with an agent (VS Code, Cline) is already covered by the Local Copilot Kit, including its team-deployment angle. This guide stays at the point-and-click "knowledge base" level of Open WebUI for one typical use case (an internal wiki, contracts), without duplicating either guide.
What is the anchoring technical profile of this guide?
A team of 15 to 20 users, a dedicated Linux server (Ubuntu Server) with a shared professional GPU in the 24-48 GB VRAM tier, Ollama + Open WebUI in Docker, authentication through your company IdP (Microsoft Entra ID walked through step by step). That headcount sits under the threshold where several enterprise-tier licensing terms in this space kick in — the most common real first deployment. Size variants (5 users, 50+, multi-site) are covered in tables in Ch. 8 and Ch. 12, without changing the step-by-step path.
How long from scoping to a running pilot?
That depends on your organization, not on this guide — a reader in a hurry who reads Ch. 1-3 (Part 1) already walks away with a memo ready for leadership, a filled-in privacy checklist and a clear read on their EU AI Act status, enough to start a real internal conversation. Chapter 11's structured pilot runs 30 days once the architecture (Part 2) and the acceptable-use policy (Ch. 10) are in place.
Is it really free of cloud cost once deployed?
The model runs on the company's own server, with no per-seat cloud subscription — Ch. 12 gives you the method and a TCO calculator to price your own case over 3 years (hardware, electricity, maintenance), without ever inventing a cloud price or an electricity bill for you. This guide itself (guide + templates + lifetime members area) is a one-time purchase of $24.
What if it doesn't work out for me after buying?
30-day money-back guarantee, one email, no questions asked — like every guide on BestLLMfor.com.
How is it delivered, and for how long?
Instant access after payment: your lifetime members area (always the current edition) plus the PDF and the template-pack zip to download, yours forever. The law (EU AI Act, privacy law) and the tools (Ollama, Open WebUI) keep moving after publication — future updates show up in your members area at no extra cost.
Can I get an invoice for accounting or an expense report?
Yes. As soon as you pay, you get an emailed Stripe receipt automatically. A formal invoice is available on request — just reply to the delivery email with your billing details.
What if I want more than one guide?
All the kits, for life bundles every current and future guide into one lifetime purchase for $49. Already own this one? The upgrade costs $25, not the full $49 — the discount applies automatically at checkout.
What does on-premise AI actually mean for a company?
On-premise AI means the model and the server it runs on sit inside infrastructure your company controls -- typically a Linux box with a GPU, on your network or a private cloud instance you fully manage. It removes the third-party processor from the equation, but your company still carries the data-processing responsibility, exactly as the guide's privacy-law chapter explains.
Is a self-hosted LLM GDPR compliant by default?
No -- running a model on your own server removes the cross-border transfer and third-party processor questions, but it doesn't automatically satisfy every other GDPR obligation. You still need a lawful basis, a retention policy, and documented security measures, which is exactly the workable checklist Chapter 2 walks you through, alongside CCPA/CPRA and HIPAA where they apply.
How many GPUs do I need for a private LLM server for my team?
There's no universal number -- it depends on your model size and how many of your 15-20 users are active at once, not a fixed per-seat ratio. The guide anchors its architecture chapter on a single shared 24-48GB professional GPU serving that whole team, with sizing variants for smaller and larger groups in the hardware and multi-site tables.
Scope your local AI deployment.
The reference guide, from the case you bring to leadership to a server running in production for your team — privacy law, the EU AI Act, architecture, security, an acceptable-use policy, a pilot, a real TCO. Today, not "someday."
Get the kit — $24