Ethos Space / July 30 update

Wait — but
test now.

Current hardware remains scarce and expensive while new higher-memory systems are approaching. Ethos should use the wait to prove its workflows, then buy into better cost efficiency.

No purchase yet30-day workload trialModel-neutral gateway
≤ $20KRetained ceiling
not a spending target
Ethos // Local Intelligence Node01 / 06
What changed since July 21

The option set widened.
The case for waiting strengthened.

JUL 22EARLY PATTERN
192 GB enters the compact-PC tier

Framework previewed a Ryzen AI Max+ PRO 495 desktop with 192 GB LPDDR5X, faster memory and an open PCIe slot for clustering. Pricing and ship date remain unannounced.

JUL 27STRONG SIGNAL
Open weights outgrew the workstation

Kimi K3 shipped as a 2.8T-parameter MoE with MXFP4 weights. The lesson is architectural: “open” does not guarantee “single-node local.” Route frontier-scale work instead of overbuying for it.

JUL 2026OPERATING SIGNAL
Agent appliances are becoming manageable

DGX Spark’s July software release improved unified-memory OOM handling and added fleet provisioning features. The category is moving from demo box toward operable infrastructure.

Framework, July 27 · Moonshot AI / Kimi K3 · NVIDIA DGX Spark release notes

Delta, not hype02 / 06
Current shortlist

Three useful shapes.
No universal winner.

MEMORY_FIRST
Mac Studio

Quiet + high bandwidth

Best fit for very large local model experiments and low-friction office use — but Apple’s current product page lists only 36–96 GB configurations, so the former 512 GB planning reference is not presently buyable from Apple.

CUDA_AGENT
DGX Spark

128 GB appliance

CUDA-native agent development, 273 GB/s memory bandwidth, 10GbE and ConnectX-7. Stronger software path; lower memory ceiling.

VALUE_LINUX
Framework

192 GB coming

More memory at the compact-PC tier with Linux and clustering intent. Interesting, but still preview-stage until price, availability and independent workload results exist.

Apple: current Mac Studio configurations · NVIDIA: DGX Spark specifications · Framework: 192 GB coming soon

Choose by workload topology03 / 06
Target architecture

Own the control plane.
Rent the frontier.

ETHOS // WORK

Private inputs

  • Technical documents
  • Research archives
  • Recurring agent jobs
POLICY // GATEWAY

One routing layer

  • Identity + permissions
  • Local-first model routing
  • Evaluation, cost + audit logs
COMPUTE // MIX

Owned + adapted

  • Local models for routine work
  • Ethos-tuned adapters
  • API/cloud for frontier work

Open weights let organizations inspect and adapt models to their own needs; lightweight LoRA adapters make that tuning more efficient.

Control + adaptation are durable assets04 / 06
Purchase gates

Buy evidence.
Not specifications.

Quality gateA local model completes at least 80% of the chosen recurring tasks at acceptable review quality.
Privacy gateThe work contains data that should remain on company-controlled infrastructure.
Utilization gateThe node has a credible path to at least 30% active utilization across scheduled and interactive work.
Economics gateThree-year total cost is favorable against the measured API and rented-compute baseline.

These thresholds are proposed decision rules, not external benchmarks. Change them before procurement if Ethos has stricter service or compliance requirements.

Measured tasks decide05 / 06
Next 30 days

Instrument the work.
Then choose the machine.

Select 10 real Ethos workflowsInclude document analysis, research synthesis, coding, archive retrieval and scheduled agents.
Run local + frontier baselinesCapture quality, latency, memory, energy, intervention rate and cost per accepted result.
Test one 128–192 GB node before buyingUse a loaner, returnable unit or short-term remote equivalent; do not infer production value from vendor demos.
Re-open procurement at the gate reviewCompare Mac, CUDA appliance and AMD/Linux options using the same traces.

Updated position: preserve the ≤$20K ceiling, spend only on the evaluation path, and keep the final hardware choice open.

Workflow traces become the specification06 / 06