Known limitations and claim locks
The prepared-machine alpha is intentionally narrow. That is what makes the proof trustworthy.
Allowed today
You may say:
- TARX has a prepared-machine managed-local alpha.
- The Always On loop can create a run, show lifecycle activity, expose evidence, persist through reload, and cancel active local work when owned by the runtime.
- The local model lane uses NVIDIA Nemotron GGUF through llama.cpp with model id
tarx-computer. - TARX is built on a staged architecture for local-first runtime, future mesh contribution, and enterprise governance.
Not allowed as live claims
Do not say:
- public download is ready
- clean-machine install is proven
- second-Mac execution is proven
- standalone/offline is proven
- mesh routing is live
- hosted burst or Supercomputer execution is live
- enterprise governance is live
- MCP memory is live
- NIM or TensorRT is in use
- TARX is a full Codex replacement
- TARX is first-million-user ready
- TARX automatically repairs every run
Current blocker classes
| Area | Status |
|---|---|
| Prepared-machine Always On core loop | Green for alpha proof |
| Visible cancel | Proven for prepared alpha |
| Clean-machine install | Locked |
| Second-Mac proof | Locked |
| Public download | Locked |
| Hosted Supercomputer | Locked |
| Enterprise governance | Locked |
Support rule
If a claim needs proof and the proof is not in the current runtime contract, do not publish it. Mark it as future architecture or private roadmap.
Next
Return to Opening the prepared-machine alpha or Request Mac Access .