THE HARD SEATS.

You do

• Deploy TARX-OS on the customer’s existing computers. Approval before action. Evidence after.
• Configure T-1, T-4, T-7, and T-70 for the room, the vehicle, or the field.
• Bring T-600 or T-K90 online when the work is physical. Hardware is capacity, not a costume.
• Write the runbook the customer can operate after you fly home.
• Hold the line on claims. Capable is not certified. Do not invent FedRAMP, SOC 2, or traction.

You are

• Someone who has shipped production systems in someone else’s building.
• Fluent in Linux, networks, GPU boxes, identity, and the ugly last mile.
• Calm in a SCIF, a factory, a clinic, or a classified-adjacent floor.
• Able to translate a private workflow into a working TARX installation in weeks, not quarters.

Bonus

• DoD, IC, or cleared-environment experience.
• Air-gapped or disconnected deployment work.
• Prior forward-deployed, field engineering, or customer-zero seats.

You do

• Own loco-manipulation stacks on T-600 and T-K90. Whole-body control that survives a real floor.
• Train, evaluate, and ship Isaac GR00T policies. On-device first. Cloud as a last resort.
• Bind robot perception and action to TARX: the agent decides, the body executes, evidence stays with the operator.
• Instrument failure. Falls, stalls, unsafe contact. You make the next run better.
• Never name unofficial partnerships, chassis vendors, or unreleased SKUs on a public surface.

You are

• A robotics software engineer who has shipped motion, not only papers.
• Deep in Isaac, ROS 2, whole-body control, imitation, or residual RL.
• Comfortable with CUDA, onboard inference, and the gap between sim and a loaded pack.
• Someone who treats a humanoid as a private computer with hands.

Bonus

• Deployed humanoid or quadruped work in the field.
• Safety monitors, contact-rich manipulation, or teleop-to-autonomy handoff.
• On-robot LLM/VLA policy work that actually closed a loop.

You do

• Design the mesh: how opted-in TARX devices contribute capacity, and how they stop.
• Own routing, rate, isolation, and audit for T-Supercomputer at https://api.tarx.com/v1.
• Keep Howdy as the developer console. No fake keys, playgrounds, or billing theater.
• Make failure loud. If a job cannot stay private, it does not run.
• Software stays free and unlimited on the computer you already own. The mesh is the paid extra mind.

You are

• A distributed-systems engineer who has built inference, routing, or multi-tenant runtimes.
• Paranoid about isolation. You think in trust boundaries, not dashboards.
• Fluent in high-throughput networking, schedulers, GPU pools, and evidence trails.
• Someone who can hold a mesh to product law: TARX is one private always-on agent.

Bonus

• Prior work on inference meshes, confidential compute, or air-gapped model serving.
• OpenAI-compatible API surfaces you would actually give a developer.
• Experience operating a console that is a real product, not a mock.