Zero Network Calls, verified
We packet-capture every build at the OS level — outbound traffic of any kind, including telemetry and license checks, fails certification.
Verified Offline · Zero Network Calls
AI that works with the cable pulled.
A curated marketplace for local-first AI — models, desktop apps, and privacy-hardened tooling that run entirely on your hardware, verified to make zero network calls.
The standard
Airplane mode is the test, not the excuse
Every listing on UnpluggedAI is installed on a clean machine, cut off at the network interface, and run again. If it degrades, throws an error, or phones home, it does not get the badge. That is the whole standard.
No account required
Buy with an email and a license key, or with crypto and no email at all; nothing in the software ever asks who you are.
One-time purchase, permanent license
Your key activates offline and stays valid for the version you bought — no renewal, no revocation, no subscription hostage.
Honest hardware ratings
Real VRAM footprint, RAM ceiling, disk size, and tokens per second measured on consumer machines we actually own.
Open weights labeled clearly
Every model shows its license, base weights, quantization, and whether commercial use is permitted — no ambiguity at checkout.
Air-gap ready downloads
Signed installers with SHA-256 checksums and offline model bundles you can carry in on a USB drive.
The Verified Offline badge
A published test report accompanies every certified listing: capture duration, interfaces monitored, DNS queries observed, and the exact build hash tested. If a vendor ships an update, the badge expires until it is re-tested.
pcap loggedDNS: 0 queriesbuild hash pinnedre-test on update
The catalog
The catalog: pick your use case, pick your hardware tier
Filter by what you need done and what you have to run it on — thin laptop, 16GB unified memory, or a 24GB GPU. Prices are shown up front, downloads start immediately, and every entry lists its size before you commit a gigabyte.
Private chat & assistants
Local chat apps with document memory, running 7B–70B models with no prompt ever leaving the box.
Offline transcription
Whisper-class speech-to-text with speaker labels — interview-grade accuracy on a plane, in a courthouse, in a clinic.
Local image generation
Diffusion pipelines with model management and batch queues, tuned for 8GB, 12GB, and 24GB cards.
Code assistants
IDE-integrated completion and refactoring against your private repos, with no source code uploaded anywhere.
Document RAG
Index thousands of contracts, records, or PDFs into a local vector store and query them in plain language.
Hardened utilities
Redaction, OCR, translation, and PDF extraction tools that operate purely on local files.
0outbound packets allowed
240+certified models & apps
8–48GBVRAM tiers covered
1×payment, permanent key
Honest comparison
Unplugged vs. cloud AI
Cloud AI rents you intelligence and keeps your data as collateral. Local AI is a capital purchase that appreciates as open models improve. Here is the honest comparison, including where cloud still wins.
Privacy
Unplugged: prompts and files never leave the device. Cloud: every input crosses a third-party network and sits in someone else's logs.
Cost over time
Unplugged: one purchase plus electricity. Cloud: a metered bill that grows with your usage, forever.
Latency
Unplugged: first token in milliseconds, no round trip, no rate limits. Cloud: fast until the API is throttled or degraded.
Uptime
Unplugged: works in a basement, a bunker, a flight, a blackout. Cloud: works when their status page says so.
Compliance
Unplugged: no data processor to vet, no cross-border transfer, no DPA. Cloud: contracts, audits, and residency questions you must answer.
Raw ceiling
Cloud frontier models are still larger. If you need the absolute maximum capability and can accept the exposure, use them — we will tell you when.
Hardware & setup hub
Hardware & setup hub
The fastest way to waste money on local AI is buying the wrong VRAM. Use our calculator to see exactly which models and quantizations fit your card, then follow build guides written for real budgets — from a $600 mini-PC to a quad-GPU inference box.
VRAM calculator
Enter your GPU and get the exact parameter count, quantization, and context length you can actually run.
Build guides
Three tested builds — Travel, Homelab, and Workstation — with parts lists, thermals, and measured tokens per second.
Quickstart docs
Ten-minute installs for Windows, macOS, and Linux, including fully offline and air-gapped install paths.
Hardware picks
GPUs, mini-PCs, and NVMe storage we've benchmarked ourselves, with notes on where the cheaper option is the right one.
“There is no such thing as a private prompt on someone else's computer. Sovereignty starts at the network interface.”
— The UnpluggedAI certification charter
Field proof
Used where uploading isn't an option
Our buyers are not hobbyists avoiding a bill. They are people whose work makes cloud AI a liability — privileged documents, patient records, classified networks, and sites with no signal at all.
"Privileged discovery never touches an outside server. Our ethics counsel signed off in a single meeting."
Litigation, 4-partner firm
"Session transcription runs on my laptop with WiFi off. There is no vendor in the room, so there's nothing to disclose."
Clinical psychologist
"Six weeks in the field, no satellite budget. The assistant worked every day exactly like it did at home."
Geologist, remote survey
"I ran the packet capture myself before trusting the badge. It was clean. That's the first time I could say that."
Security engineer, homelab
Air-gapped enterprise
Site licenses, offline activation, signed media delivery, and support tiers for defense, finance, and critical infrastructure.
Small IT teams
Volume keys, internal deployment scripts, and compatibility reports so you can roll out AI without opening a firewall rule.
Membership & vendors
Go Pro, or list your model
Pro membership unlocks the full model library, version updates, the hardware compatibility tooling, and early access to newly certified releases — one flat annual fee, no per-token anything. Vendors: if your app genuinely runs offline, we will prove it and put it in front of buyers who pay for exactly that.
Pro membership
$149 / year
Full library access, update channel, VRAM tooling, and priority certification requests.
Join Pro →List your model or app
Rev share
Revenue share on sales, plus optional Verified Offline certification with a published test report.
Apply as a vendor →Enterprise & air-gapped licensing
Custom
Site-wide keys, offline activation servers, signed physical media, and named support engineers.
Talk to licensing →The dispatch
Free
A short monthly email: newly certified local models, benchmark updates, and hardware price drops worth acting on.