The Hardware

Four Machines.
Four Missions.

From palm-sized AI nodes to discrete-GPU workstations, every machine in the lineup is purpose-built for local AI, edge compute, and sovereign data processing. No cloud required.

We can ship any of these devices to you with OpenClaw pre-installed.

Hala Nexus (STHT1) main view
Hala Nexus (STHT1) alternate view
The AI Inference King

Hala Nexus(STHT1)

Up to 128 GB of unified LPDDR5 memory lets you run 235-billion-parameter models locally. The definitive flagship with AMD Strix Halo AI 395.

Run 235B parameter models (2-bit quantization)
Flagship 16-core Strix Halo architecture
80 TOPS NPU for massive local inference

CPU

Strix Halo AI 395 (16C)

Memory

128 GB LPDDR5

AI / NPU

80 TOPS NPU

GPU

40CU RDNA 3.5

Free shipping included

Available with OpenClaw pre-installed — we can send it to you ready to run.

Hala Pulse (STHT1) main view
Hala Pulse (STHT1) alternate view
Ultra-Compact Powerhouse

Hala Pulse(STHT1)

The same 16-core Strix Halo power as the Nexus, in a sleek 127mm metal chassis. 128GB RAM for unrestricted local AI compute.

128GB LPDDR5 (8533MT/s) in a palm-sized box
80 TOPS NPU + 40CU RDNA 3.5 Integrated Graphics
Compact, metal build with superior thermal management

CPU

Strix Halo AI 395 (16C)

Memory

128 GB LPDDR5

AI / NPU

80 TOPS NPU

Volume

1.3L (127×127mm)

Free shipping included

Available with OpenClaw pre-installed — we can send it to you ready to run.

Hala Apex (STHT1) main view
Hala Apex (STHT1) alternate view
The High-Performance Box

Hala Apex(STHT1)

A high-authority AI workstation with 12-core Strix Halo AI 385. Professional-grade performance in a compact, mesh-cooled chassis.

12-core high-performance Strix Halo variant
80 TOPS NPU for accelerated AI workloads
Compact mesh design for optimal cooling

CPU

Strix Halo AI 385 (12C)

Memory

Up to 64 GB DDR5

AI / NPU

80 TOPS NPU

GPU

32CU RDNA 3.5

Free shipping included

Available with OpenClaw pre-installed — we can send it to you ready to run.

Hala Core (STHT1) main view
Hala Core (STHT1) alternate view
The Entry-Level AI Node

Hala Core(STHT1)

Secure, local AI for every desk. AMD Ryzen 7 7840 power in a striking vertical mesh tower. The gateway to AI sovereignty.

Ryzen 7 7840 performance for business automation
Vertical mesh tower for maximum airflow
Perfect entry point for secure, on-premise AI

CPU

AMD Ryzen 7 7840 (8C)

Memory

32 GB LPDDR5

AI / NPU

16 TOPS NPU

GPU

12CU RDNA Graphics

Free shipping included

Available with OpenClaw pre-installed — we can send it to you ready to run.

Open-source AI Agent Runtime
$ open-claw run --local --model llama3:70b

Your Device Runs Open CLAW.
AI Agents, Fully Offline.

Open CLAW is the open-source AI agent framework built for local hardware. Pair it with any Hala machine and you get autonomous agents that write code, conduct research, and complete complex workflows — without a single API call leaving your building. We can send any of our devices to you with OpenClaw pre-installed.

Autonomous Coding Agents

Let Open CLAW write, review, and ship code without touching a remote API. Full IDE-level agents running on your own silicon.

Multi-Agent Orchestration

Chain specialised agents — researcher, planner, executor — into pipelines that tackle complex, long-horizon tasks end to end.

Total Data Sovereignty

Every prompt and every output stays on your device. No telemetry. No cloud handoff. Built for regulated industries and sensitive work.

No Limits. No Subscriptions.

No API rate caps. No token quotas. No monthly bills. Run agents at full throughput, 24/7 — powered entirely by your Hala hardware.

235BParametersMax model size on Hala Nexus
0API CallsEverything runs locally, always
24 / 7Agent uptimeNo rate limits, no quota resets
Also works with:OpenHandsLM StudioOllamaJanLlamaFileAnythingLLMCrewAIAutoGen

One Ecosystem, Zero Overlap

Each machine serves a distinct purpose. Together they cover every deployment scenario from desk to data centre.

Every Workload Covered

From 128 GB unified memory for massive LLMs to a discrete GPU for real-time rendering — there's a machine for every job.

0.8L to 6L — Pick Your Size

The AKB56 fits in your palm. The XB35 houses a full desktop GPU. Choose the form factor that matches your deployment.

Whisper-Quiet to Triple-Fan

35W silent mode on the xN88, vapour chamber on the AKB56, triple-fan tower on the XB35. Thermal design matched to workload.

External or Internal Power

Compact adapters for the portable units, a built-in 336W ATX supply for the XB35. No compromises on power delivery.

At a Glance

Side-by-side, so you can pick the right machine in seconds.

SpecHala Nexus(STHT1)The AI Inference KingHala Pulse(STHT1)Ultra-Compact PowerhouseHala Apex(STHT1)The High-Performance BoxHala Core(STHT1)The Entry-Level AI Node
Best ForMassive LLM InferenceFlagship PortabilityHigh-Performance AI WorkOffice & Edge AI
CPURyzen 7 7840 (8C)
NPU / AI80 TOPS80 TOPS80 TOPS16 TOPS
GPU40CU RDNA 3.540CU RDNA 3.532CU RDNA 3.512CU RDNA
Max Memory128 GB LPDDR5128 GB LPDDR564 GB DDR532 GB LPDDR5
Volume~2 L~1.3 L~2.5 L~4 L
Power240W external230W external240W external120W external
Starting AtInquireInquireInquireInquire

All models ship with Windows 11 · 2× M.2 NVMe slots · WiFi · Bluetooth · 2.5G+ Ethernet