Private by default
Your prompts and files stay on the device in normal use. If you choose preload, it is handled as a separate, clearly defined step.
Local first · no hidden cloud dependency
Proof, not promise
Before you buy.
You should be able to see, in plain terms, why this system is worth trusting: where your data stays, what the device is genuinely good at, and what support looks like after checkout.
Capable enough to be useful
selbsai is configured around open-model classes that are already strong for drafting, retrieval, multilingual work, and document-heavy workflows.
7B–30B classes · model-fit over hype
A real delivery promise
You get human support, tracked shipping, and a visible customer portal from order to delivery.
Support · order tracking · portal
Two ways to use AI.
Only one is yours.
The cloud is rented intelligence — convenient, but not really yours. Selbs AI is owned intelligence. Here's the trade, in plain language.
Cloud AI
Rented
ChatGPT, Gemini, Claude — you ask, the answer travels through the internet to a stranger's server, and back.
What you pay
Per-question, every month
Where your data lives
On someone else's servers
How fast it answers
120 – 900 ms over the internet
If the internet goes down
It stops working
Who can revoke access
The provider, at any time
Selbs AI
On-premise
Owned
A small, quiet computer that lives in your home or office and does the thinking right there. Nothing leaves the room.
What you pay
A predictable purchase, like a fridge — not a subscription that grows with use.
Once. Then it's yours.
Where your data lives
On the device, in your home
How fast it answers
< 5 ms — local
If the internet goes down
Keeps working
Who can revoke access
Nobody but you
Why ownership mattersOwn the engine.
Private. Local. Yours.
Own the engine.
Keep the context.
If AI helps run your work, your life and business deserve more than rented access in the cloud. Local hardware now makes that level of privacy, autonomy and control practical and affordable.
Privacy by default
Prompts, files and internal context stay inside your own perimeter instead of travelling to a provider far away.
Autonomy you can feel
No remote account dependency, no provider switch to flip, and no rented intelligence sitting between you and your work.
Affordable enough to own
Serious local AI is finally practical as hardware: a one-time system you control, not a monthly model rent that keeps growing.
Cloud AI asks for ongoing trust. Owned hardware lets you set the rule.
Everyday AI work
Everyday AI, local.
selbsai is not only a model box. It is a private work surface for the tasks teams already give to AI: drafting, reviewing, searching, comparing, summarising, and preparing the next action.
Build your workflow →Local by design
Software coding
Explain code, draft tests, review pull requests, write scripts, and search local repository notes without exposing proprietary source.
Local by design
Documents and text
Draft letters, policies, proposals, reports, summaries, tables, and structured extracts from private source files.
Local by design
Email and personal assistant
Prepare replies, classify inboxes, extract action items, create agendas, and turn meeting notes into follow-up work.
Local by design
Research desk
Compare PDFs, summarize long materials, build briefing notes, and answer questions with citations from local documents.
Local by design
Sales assistant
Prepare call notes, objection handling, proposal drafts, CRM-style summaries, and account research from approved company material.
Local by design
Compliance management
Search policies, flag missing evidence, prepare audit answers, compare obligations, and keep sensitive control documents local.
Local by design
Inventory and warehouse
Summarize stock lists, surface reorder issues, draft supplier emails, review SOPs, and answer operations questions from local records.
Local by design
Document review
Find clauses, obligations, inconsistencies, deadlines, unusual terms, and missing attachments across contracts and case folders.
Local AI is getting faster
Local AI is catching up.
The open-model ecosystem is moving from raw model releases to runtime acceleration: speculative decoding, multi-token prediction drafters, MLX, Ollama, vLLM, SGLang, quantization, and hardware-specific tuning. A selbsai device is built to benefit from that curve without moving your work back into the cloud.
Google's Gemma 4 MTP drafters are a clear signal: open models can gain major responsiveness improvements while the main model still verifies the output.
Read the Gemma 4 MTP note →Faster daily chat
Drafting, email review, document Q&A, and personal assistant flows feel better when token latency drops.
Better local agents
Coding helpers, research loops, and multi-step workflows benefit when each step returns faster.
Future model swaps
The device is not tied to one model generation. Better open models and runtimes can be adopted over time.
Benchmarks without the theatre
Benchmarks change.
The selbsai promise is not that a small local model beats every frontier API. The promise is that many daily business tasks do not need the frontier API at all: drafting, coding support, document review, internal search, email work, compliance workflows, and private knowledge assistance can run on a device you control.
- Use premier cloud only when you truly need maximum frontier reasoning.
- Use local open-weight models for private, repeated, document-heavy workflows.
- Switch models and runtimes over time without rebuilding the customer workflow.
Source
Live model leaderboard
For current rankings, use Artificial Analysis directly. Their leaderboard updates as providers release new frontier, open-weight, speed, quality, and pricing data.
Artificial Analysis Intelligence Index
Benchmark results move quickly and local speed depends on hardware, quantization, model choice, and acceleration paths such as MTP drafters, MLX, Ollama, vLLM, and SGLang.
Open model ecosystem
Curated open AI.
There are far more models than a normal buyer should have to compare. selbsai filters the ecosystem by workload, source reputation, license posture, runtime, format, hardware fit, and update policy before a system is provisioned.
Model-card review
We prefer models with clear intended use, known limits, benchmark context, and maintained release history.
Runtime and format fit
GGUF, Safetensors, MLX, Ollama, llama.cpp, and vLLM choices are matched to the device and the customer workflow.
Update posture
Customers can choose stable, balanced, or fast-track model updates instead of waking up to surprise behavior changes.
Pick your headache
Choose the job.
Tell us the problem in plain language. We'll show you the unit and workflows that solve it — and pre-fill the configurator.
Research desk
Drop them in. Ask anything. Get answers with citations, instantly. None of it leaves the device.
Configuration
Obsidian Core · Professional · 2 TB · Research desk + Knowledge base
From box to brain
Three minutes, three steps. Nothing technical.
If you can set up a wireless printer, you can set up Selbs AI. If you can't set up a wireless printer — Selbs AI is even easier.
- 01Plug
Power and ethernet.
It boots in under 30 seconds. No installation wizards, no driver hunts.
- 02Connect
Open selbsai.local on your laptop.
Any device on your network sees it instantly. No accounts, no cloud sign-in.
- 03Command
Start asking questions.
It’s already loaded with a tested model and your chosen workflows. From box to first answer in under three minutes.
Your browser barHTTP · local
http://selbsai.local
That's it. The unit advertises itself on your local network via mDNS. No port-forwarding, no certificates, no DNS records to touch.
Daily control
Chat. Control. Export.
After setup, selbsai is used through a private local browser interface. The customer decides the workflow, the document scope, and what happens to the answer.
See use cases →Local browserselbsai.local
Prompt
Review this supplier contract and list payment risks with source references.
Draft answer ready. Sources attached. Export, revise, or continue in chat.
Scope: Contracts / Supplier A
Mode: Document review
Output: Draft + citations
Chat
Ask from any browser
Open the local interface on your laptop, desktop, tablet, or phone and chat with the device like a normal AI assistant.
Scope
Choose what it can read
Point a question at a folder, document set, or workflow so answers come from the right local material instead of every file at once.
Mode
Switch workflows and models
Use prepared modes for coding, drafting, research, review, sales, compliance, warehouse, or general assistant work.
Output
Keep answers usable
Copy, revise, cite sources, turn answers into drafts, and keep sensitive outputs inside the local workspace until you decide otherwise.
What ships
What ships.
Confidence increases when the offer is concrete. Every node should read like a complete system, not a bare mini-PC with vague AI claims attached.
Configured hardware node
A selected device, tested memory profile, storage, and the right open-model class for the chosen workload.
Ready local stack
Inference runtime, retrieval tooling, task presets, and optional workflows installed before delivery.
Onboarding and portal access
A customer portal, upload path when needed, and clear next steps from order to first answer.
Support path
Human support, customer portal access, and a known route to help if setup or operation needs attention.
Where your data goesOne trip across the room.
One trip across the room.
Versus seven across the world.
Below is the actual path your question takes — with a typical cloud assistant, and with Selbs AI. It's not a metaphor. It's literally what happens, every time.
Cloud assistant
Cloud · transit
~ 4.5 s
- Your question→
- Your router→
- Internet exchange (Frankfurt → New York)→
- Provider data centre→
- Logged & analysed→
- Back across the Atlantic→
- Back to you
Selbs AI
Local · private
~ < 1 s
- Your question→
- The box under your desk→
- Back to you
If the internet cable is cut, Selbs AI still works.
Can the cloud say that?
Savings calculator
Pay once.
Most teams break even on Selbs AI in under 18 months. After that, the intelligence is essentially free. Move the sliders to see your own numbers.
Your team today
5 people
25 € / month
Monthly cloud cost€125.00
Cloud cost over 36 mo.€4,500.00
Break-even — versus Selbs AI
16months
16 months, your Selbs AI pays for itself. After that, your intelligence is free forever.
Cloud over 36 mo.€4,500.00
Selbs AI · one-time€1,977.74
Reference buildObsidian Core · Professional · 2 TB
36-month savings€2,522.26
Estimates only. Numbers exclude electricity (~ 1–2 € / month) and assume cloud subscriptions are paid per active seat.
European Assurance
Promises in writing.
Trust is the highest currency for a product like this. Three things we commit to in writing for every customer.
01In writing
No hidden backdoors.
We use open-weight models and publish the firmware build. Anyone — your IT team, an auditor, a curious neighbour — can verify what runs.
Open weights · reproducible firmware · published SBOM
02In writing
A physical kill-switch.
A real switch on the chassis cuts the network entirely. Air-gap mode is one click away — not a setting buried in a menu.
Hardware switch · LED-confirmed · no software override
03In writing
Local, human support.
Made in Europe. If something breaks, you talk to one of the engineers who built it. No call-centre, no AI chatbot, no ticket queue.
Mon – Fri · 09:00 – 18:00 CET · DE / EN
After you order
After checkout.
A premium hardware purchase needs post-purchase clarity. The buyer should know exactly what happens next and when support enters the picture.
- 01
Checkout and confirmation
The order is confirmed, payment status is visible, and the customer portal becomes the control point.
- 02
Optional preload intake
If selected, the customer receives a secure upload route and clear handling expectations for private files.
- 03
Provisioning and QA
The node is configured, tested, and prepared around the selected workload before it leaves the bench.
- 04
Shipping and onboarding
Tracking, expected delivery, and onboarding guidance should remove ambiguity from the final handoff.
Use-case appliances
Built for your work.
Each selbsai build starts with the profession, the documents, the risk level, and the operating boundary. The model is only one part of the appliance.
Configure an appliance →Lawyer
Typical workload
Contracts, case files, client memos
What on-premise AI can do
- Turn every matter folder into a private, cited second brain
- Draft first-pass letters, clauses, and timelines from your own files
- Ask sensitive case questions without risking privilege in a cloud prompt
Why this beats cloud
Cloud AI can turn a simple drafting task into a privilege, vendor, and data-transfer discussion.
Doctor
Typical workload
Letters, notes, patient-facing summaries
What on-premise AI can do
- Convert notes into clear patient letters before the next appointment
- Find the right form, wording, or guideline while the patient is still in mind
- Reduce evening paperwork without sending practice data to a model provider
Why this beats cloud
Cloud AI may save minutes, then cost hours explaining patient-data handling and access controls.
Steuerberater
Typical workload
Receipts, invoices, DATEV-oriented exports
What on-premise AI can do
- Find the missing receipt, wrong total, or odd booking before deadline week
- Extract invoice facts into review-ready tables instead of retyping them
- Answer client questions from their own file history while staying audit-ready
Why this beats cloud
Generic cloud AI is not built around German tax confidentiality, DATEV habits, or client-file minimisation.
Architect
Typical workload
Specifications, tenders, site notes
What on-premise AI can do
- Ask a project what changed, what is missing, and where the risk sits
- Draft site reports, RFI answers, and tender summaries from local files
- Keep drawings, specs, and bid material out of public AI workspaces
Why this beats cloud
Cloud AI can make unpublished bids, designs, and client plans feel like material you no longer fully control.
Agency
Typical workload
Brand voice, creative drafts, client research
What on-premise AI can do
- Keep every client voice, campaign angle, and rejected idea searchable
- Create concepts, variants, and research packs before the pitch call
- Work on embargoed launches without training someone else's future tool
Why this beats cloud
Cloud AI is fast, but it can blur ownership around prompts, drafts, client strategy, and pre-launch work.
Investment, insurance & pension advisors
Typical workload
Suitability notes, product documents, client portfolios
What on-premise AI can do
- Prepare sharper client reviews from policies, portfolios, and pension files
- Find the clause, risk, fee, or suitability note while the client waits
- Keep wealth, health, and family financial context inside your advisory office
Why this beats cloud
Cloud AI can create needless exposure around wealth, insurance, pension, and suitability records.
Works council
Typical workload
Policies, minutes, employee questions
What on-premise AI can do
- Turn agreements, policies, and minutes into a searchable employee-rights memory
- Prepare meeting positions from the documents everyone argues about
- Handle sensitive workplace questions without pushing employee data into SaaS AI
Why this beats cloud
Cloud AI is hard to defend when employee data, monitoring concerns, and internal conflicts are in scope.
Real estate advisors & property managers
Typical workload
Leases, exposés, owner reports, due-diligence files
What on-premise AI can do
- Ask any property folder for rent terms, obligations, deadlines, and open issues
- Draft owner updates, tenant replies, and deal summaries in minutes
- Review due-diligence packs without leaking pricing or transaction context
Why this beats cloud
Cloud AI can expose tenant, owner, pricing, and transaction context in a business built on discretion.
Buyer questions
Before you buy.
Can this run without the cloud?
Yes. The core experience is designed to work locally on the device.
Do I need to understand Linux or model hosting?
No. The point is to receive a configured node, not a hobby project.
What kinds of tasks is this actually good for?
Drafting, retrieval, multilingual work, document-heavy workflows, and private assistants are the core fit.
What if I need help after delivery?
The customer portal, support path, and onboarding materials are part of the product experience.
Made in
Europe
Standard
DSGVO-konform
Tax
MwSt-konform
Shipping
DHL · DE / AT / CH