The market sells a chat UI inside your network. TARX_OS is the always-on runtime the company already locks. Join on computers you have. Get a T-4 or T-70 when you want dedicated NVIDIA.
CIO / CISO / platform owners who already lock the computers and will not put the company’s loop on a rented API.
The company already has computers it secures. Put the runtime on them.
TARX software
Free
Not a consumer subscription.
Typical 5-person mix
$280/mo
What a chat stack actually costs before rack.
T-70 in 1U
$444/mo
$16,000 list ÷ 36.
Category in 2026
Chat in a VPC
On-prem is still sold as a window with an OpenAI-compatible API.
THE CATEGORY
What they sell. What we ship.
The market
Enterprise “private AI” is a chat UI inside your network. Policy is a slide. Evidence is a vendor log.
TARX
TARX_OS is the governed runtime the company already locks. Join on computers you have. Own a T-4 at the desk or a T-70 in the rack.
Who this page is for
CIOCISOHead of platformIT operationsProcurement
What this buyer is already hiring for
On-prem / air-gapped LLM
MLOps that cannot leave the VPC
AI governance / agent approval
Inference capacity the CIO can point at
HOW TARX WORKS
How TARX works in the enterprise
01
Put TARX_OS on existing silicon
Join. free software. First-run stays local.
02
Govern the loop
Policy. Deny or Allow. Evidence after every write.
03
Desk capacity
T-4 when a department needs dedicated NVIDIA beside the team.
04
Rack capacity
T-70 in 1U when the room needs it. Same OS. Same contract.
The company already has computers it secures. Put TARX_OS on them. That is Join. Not a cloned chat window with an OpenAI-compatible API. Always-on TARX under Control. Evidence after every write.
Typical Claude mix is 80% $20 and 20% $200, whole seats. Five people: $280/mo. TARX software is free. When a department needs dedicated NVIDIA, get a T-4. When the room needs rack capacity, get a T-70 in 1U. Hardware monthly is list ÷ 36.
The CIO can point at one contract on Macs, at the desk, and in the rack. First-run stays local. Join the Supercomputer. Or own Private Hardware.
THE WORK
WHAT ENTERPRISE ACTUALLY RUNS.
Department work that cannot hit a rented API
Prompts, files, and weights stay inside the company unless you approve a route.
A private cluster the CIO can point at
Macs already running TARX_OS. T-4 beside the team. T-70 in the rack they already lock.
Audit after the action
What ran, what changed, who approved. Same contract on TARX_OS and Supercomputer.
ECONOMICS
Seats versus a runtime the company owns.
Path
Monthly
Contract
Typical 5-person chat mix
$280
Rented seats. 4 × $20 + 1 × $200.
TARX_OS
$0
Runtime on computers you already secure.
T-4
$183
Indoor / vehicle Black Box.
T-70
$444
1U rack. List ÷ 36.
The CIO can point at one contract on Macs, at the desk, and in the rack.
THE STACK
Join the Supercomputer. Or own Private Hardware.
Operating state, policy, and evidence on owned computers. T-4 at the desk. T-70 for 1U rack capacity in a room you already secure.
Placeholder
TARX_OS · Conversation into the runtime · free software
Model pathTARX_OS on Mac. T-Supercomputer on the mesh.
GovernancePolicy. Approval. Evidence.
Is this a new SaaS seat?
No. TARX software is free. Own Private Hardware when you want dedicated NVIDIA on the premise. T-Supercomputer is paid external inference, not a consumer subscription.
How does that compare to a typical Claude mix?
Typical mix is 80% $20 Pro and 20% $200 Max, whole seats. Five people: 4 × $20 + 1 × $200 = $280/mo. T-4 is $6,600 ÷ 36 = $183/mo. T-70 is $16,000 ÷ 36 = $444/mo.
Where does the data go?
It stays on the premise unless you approve a route. Air-gapped capable.
TWO EQUAL PATHS
JOIN THE SUPERCOMPUTER. OR OWN PRIVATE HARDWARE.
Join the Supercomputer. Or own Private Hardware.
JOIN · TARX_OS
Join the Supercomputer
Give TARX superpowers.
TARX works privately on your computer. Connect to the Supercomputer when you want TARX to research the web, use more skills, and take on bigger jobs with deeper models. Supercomputer connection is in beta, and capacity is being qualified.
