Viewport 02 · Architecture

Seven layers, one boundary

Everything from L0 to L6 runs on the client's premises. The frontier layer is internal ShurIQ R&D only, it never crosses to the client side.

Client premises, runs totally locally

Hardware

Mac mini M4 Pro · RTX 5090 · DGX Spark · Mac Studio M3 Ultra · RTX PRO 6000 · H100 node

client-owned

Inference runtime

Osaurus (single-analyst) / vLLM (multi-seat), local/cloud endpoint swap, no code change

MIT / Apache

Base model, frozen

Qwen 3.5 · DeepSeek V3.2/V4 · Mistral · gpt-oss, at Q4_K_M minimum

Apache / MIT

Parametric injection

DMoE LoRA expert bank, per-client corpus + vertical priors + SBPI rubric

DMoE method

Retrieval / orchestration

Agents-K1 · GraphRAG over Oxigraph SPARQL · BM25 · mem0

Agents-K1 / Oxigraph

Knowledge graph

RDF triple store (105,738 → ~1B facts), SBPI ontology, full graph at ShurIQ, slice on-prem

Oxigraph / InfraNodus

Agent harness

Report Engine · SBPI scorer · fact-check ledger, behind a local endpoint

ShurIQ harness

Never crossesConfidential client research stays below this line. Frontier models stay above it.

ShurIQ internal, R&D only, never client-facing

R&D

Frontier models

Claude Opus 4.8 / Sonnet 4.6, graph construction, rubric training, higher-level synthesis. Output is an asset (the injected KG) shipped to the client model.

ShurIQ infra