← LoLR HPCs
LoLR BOXCore product

Materials, knowledge, and workflows
Reside on your own desktop device

Designed for tax professionals, finance teams, lawyers, investment researchers, risk managers, consultants, and researchers. Its heterogeneous dual engines—Moore Threads E300 SoC and Houmo M50 accelerator—support local models, private knowledge bases, and professional Skills in a compact desktop device.

Product modelAISC-B100-M50E300 + M50 current main configuration
AI computing powerUp to 210 TOPS · INT8E300 50 + M50 160 combined physical peak; not a unified compute pool
Memory32 GB LPDDR5 + 24 GB card memoryHost and accelerator memory are independent
Dimensions201 × 128 × 95 mmD × H × W
View specifications

Design philosophy

LoLR BOX is an on-device AI space that continuously holds your professional materials, methods, and workflows, rather than a general-purpose chat device. Sensitive tasks can prioritize local parsing, retrieval, and generation. Cloud models or external tools are enabled by user choice, with data boundaries governed by deployment policy. AI outputs support judgment and require final review by professionals.

LoLR BOX
Product images are for identification, not dimensional reference. Appearance is subject to official release and actual delivery.

Users & Scenarios

  • Tax, Accounting & AuditPolicy searches, draft filing plans, and risk self-checks.
  • Finance & Risk managementFinancial report organization, compliance reviews, and due diligence materials.
  • Lawyers & Legal teamsContract and case-file analysis, key-point extraction, and risk alerts.
  • Research & Small teamsPrivate knowledge bases, literature review, and draft reports.

Features & Value

Professional scenario support

Load knowledge, rules, and Skills for tax, accounting, law, finance, and research tasks.

Local-first processing

Common parsing, retrieval, and generation tasks can run on the device.

Build private knowledge

Project materials, conversations, and reviewed results accumulate in your own space.

Controlled device–cloud collaboration

Invoke external capabilities according to task authorization and deployment policy.

Technical specifications

The table reflects the current product definition. Computing power is a theoretical peak; actual speed depends on the model and task.

ItemDescription
Product modelAISC-B100-M50E300 + M50 current main configuration; official part number, board form factor, and driver version pending confirmation
Computing platformMoore Threads E300 SoC + Houmo M50E300: 8 performance cores, up to 2.65 GHz per core; M50 is an independent AI accelerator
AI computing powerUp to 210 TOPS · INT8E300 50 + M50 160 combined physical peak; not a unified compute pool
M50 capabilities160 TOPS INT8 · 100 TFLOPS bFP1624 GB dedicated accelerator memory; not a measured performance result for a specific model
Memory32 GB LPDDR5 · 102.4 GB/s + 24 GB accelerator memoryHost and accelerator memory are separate, not a unified 56 GB pool
Local storage1 TB NVMe · PCIe 4.0 ×4M.2 2280; SKU and expansion options are subject to the delivery list
Total system powerUnder 120 WSpecific configuration, workload, and test conditions pending confirmation
Dimensions201 × 128 × 95 mmD × H × W; website image proportions are not dimensional references
About this page| Computing power is the combined theoretical peak of the main processor and accelerator. Host and accelerator memory are independent; actual speed depends on the model and task.