Onboarding a team that only knows ChatGPT
Show LOCAL, approval, and Evidence on one real file. Then talk economics.
← Answers
education · today · What should every team know about private AI in 2026?
Literacy before another seat.
Four splits. Chat versus a runtime that moves work. Local versus a rented API. Memory versus Vault. Join versus own. If a team cannot draw those lines, another $20–$200 seat will not make them literate. TARX-OS makes the lines visible. First-run stays local. Approval before a write. Evidence after.
Literacy is operational. If you cannot see what ran, you do not govern it.
TARX-LOCAL handles private work on your computer. Connections stay off until you allow them.
Human funnel: People →
HOW TARX WORKS
How TARX works as the literacy layer
A chatbot returns text. An agentic runtime accepts a goal, calls approved skills, waits when it must, and leaves Evidence. That is the first split. Most tools still sell the first as if it were the second.
The second split is where the loop runs. TARX-OS runs on Apple silicon you already own. Nothing attaches until you opt in. Air-gapped capable. Web lookup is labeled and off until you say so. Cloud seats cannot teach that, because the product is the hop.
The third split is Memory versus Vault. Context is not a password. The fourth is Join versus own: $0 TARX-OS seats on computers you have, or T-1 / T-4 / T-7 / T-70 when you want dedicated NVIDIA. Literacy is being able to choose.
THE STACK
Join the Supercomputer. Or own Private Hardware.
TARX-OS on computers you already own. Supercomputer hardware when the site needs dedicated NVIDIA.
USE CASES
WHAT THIS PAGE IS FOR.
A security review that needs plain language
Memory is context. Vault is secrets. Writes wait. Search can stay off.
A budget meeting
Typical Claude mix is 80% $20 and 20% $200, whole seats. TARX-OS seats are $0. Hardware is list ÷ 36.
KEY FEATURES
Dialed to this question.
Chat vs runtimeConversation is the interface. The runtime is the product.
Local defaultFirst-run stays local. Nothing attaches until you opt in.
ControlDeny or Allow before a consequential write.
EvidenceWhat ran, what changed, who approved.
PathsJoin the Supercomputer, or get your own.
HOW TO START
Four steps. Same contract.
01
Name the four splits
Runtime, locality, secrets, ownership.
02
Run them on TARX-OS
Do not lecture them on a whiteboard and send them back to a seat.
03
Keep search off until it is a feature
Air-gapped capable is a teaching tool, not a slogan.
04
Choose Join or own
Macs you have, or TARX hardware. Same OS.
FAQ
Questions this page owns.
What is the one-sentence literacy test?
Can the team say where the loop ran, who approved the write, and whether secrets mixed into memory?
Do we need a T-70 to start learning?
No. TARX-OS on existing Apple silicon is the start. Supercomputer is optional.
Is Web lookup required for literacy?
No. It is labeled, Evidence-backed, and off until you opt in. Keeping it off is a valid lesson.
How is this different from an LMS module?
An LMS describes AI. TARX-OS is AI under your policy. The product is the lesson.
TWO EQUAL PATHSJOIN THE SUPERCOMPUTER.
JOIN THE SUPERCOMPUTER.
OR OWN PRIVATE HARDWARE.
Join the Supercomputer. Or own Private Hardware.
JOIN · TARX-OS
Join the Supercomputer
Always-on agents on silicon you own.
TARX is not a chat app. TARX-OS is the always-on governed runtime for AI agents and work. Install it on the Mac you already have. Chat is how you talk to it. The runtime stays on — goals, skills, approval, Evidence.
Join beta →OWN · TARX HARDWARE
Own Private Hardware.
When local is not enough, attach a TARX machine.
T-1, T-4, T-7, T-70, T-600. Model numbers. They join the Supercomputer — the mesh — for private cluster, burst, or air-gapped sites. Same OS. Same approval. Same Evidence.
Explore Private Hardware →DEPLOYMENT DESIGN · T-4OWN YOUR COMPUTE.
OWN YOUR COMPUTE.
DESIGN THE DEPLOYMENT.
TARX-OS seats are included. Configure the Supercomputer. Watch the math. Submit once.