Skip to Content

What Always On does

Always On is the current alpha surface for the local execution loop. It is the founder-open default for prepared-machine review.

Current job

Always On lets a reviewer:

  • submit a prompt into the local runtime
  • create a run
  • see lifecycle activity from the runtime
  • inspect evidence surfaces
  • reload and find the selected run again
  • cancel an active run through the visible cancel control

Why it matters

Always On is not a generic chat surface. It is the review surface for the core TARX loop:

intent -> run -> lifecycle -> evidence -> persist

Future product language may include plan, act, check, repair, evidence, and persist. For this alpha, do not claim plan/check/repair unless the specific behavior is shown by the product.

Current model path

The prepared alpha uses a local model lane:

FieldValue
Runtime kindllama_cpp_gguf_nvidia_nemotron
Providernvidia_local
Endpointhttp://127.0.0.1:11443
Model idtarx-computer

This is not NIM, TensorRT, hosted inference, or Supercomputer execution.

Support note

If Always On opens to an older demo or product-path session, treat that as a support issue. The accepted alpha default is TARX Always On.

Next

Continue to The run lifecycle.

Agents