NOĒSIS
Patent-Pending · Published on arXiv · by Alpha Cogs

Your knowledge is scattered.
Noēsis connects it.

The first Graph-RAG system that beats GPT-4o on academic benchmarks — running on a single consumer GPU. No cloud required. No data leaves your infrastructure.

+28
points above GraphRAG
23×
faster ingestion
96%
of SOTA accuracy
<2ms
cross-KB routing
📄 Published on arXiv 🔒 Patent Pending (IT) 🎤 Presented at IAMT UK @ AWS HQ 🏆 HotpotQA Validated

Proven Performance

Validated on HotpotQA — the standard multi-hop reasoning benchmark. 1,000 questions. No cherry-picking.

SystemEM ScoreF1 ScoreInfrastructure
Microsoft GraphRAG31.7042.74GPT-4o (cloud)
BGE Dense + GPT-4o47.6060.36GPT-4o (cloud)
Noēsis59.5074.7435B model (on-premises)
HopRAG62.0076.06GPT-4o (cloud)
StepChain (SOTA)66.7079.50GPT-4o (cloud)

Noēsis uses a 35B on-premises model for graph construction. All competitors use GPT-4o for everything. Noēsis retrieves k=10 chunks; baselines use k=20. Source: arXiv:2608.15919

"A 2.3B smartphone-sized model with Noēsis architecture matches GPT-4o dense retrieval. The architecture does ~80% of the work."

— From the paper, Ablation study §5.1

Why Not Just Use GraphRAG?

We built Noēsis because existing solutions weren't good enough for production.

Traditional Graph-RAG

  • Static chunking — loses cross-section connections
  • Requires GPT-4o (cloud, expensive, data leaves)
  • Fixed parallelism — OOM crashes or wasted GPU
  • Single KB only — no cross-domain discovery
  • Hours to index medium corpora
  • No adaptation to your specific domain

Noēsis

  • Bidirectional traversal — maintains context across 300+ pages
  • Runs on 12GB consumer GPU, fully on-premises
  • AIMD adaptive — zero crashes, self-regulating
  • Cross-KB Mesh — discovers connections between domains
  • 13.4MB corpus in 66 seconds (23× faster)
  • Domain-aware optimization — adapts to your field

Four Patent-Pending Algorithms

Each solving a problem no one else has addressed.

① Bidirectional Graph Traversal

Simulates human reading with degrading memory. Forward pass builds context; backward pass reconnects early concepts to late ones. Result: 100% more edges than single-pass extraction on 300+ page documents.

② AIMD Concurrency Controller

TCP congestion control (1988) transferred to document orchestration. Probes GPU capacity in real-time. 23× faster than sequential. Zero OOM crashes — including 160+ min on 6GB GPU.

③ Moēsis (MoE Optimization)

Domain-aware selective quantization for Mixture-of-Experts models. Hot layers keep precision, cold layers compress. 6.3× prompt speedup on 12GB. Re-adapts when your domain changes — no cumulative precision loss.

④ Mesh (Cross-KB Discovery)

Multiple KBs stay separate. At query time, Mesh discovers emergent connections between them — relationships that exist in no single document. Adaptive threshold, <2ms routing, real-time structural discovery.

Built For Knowledge-Intensive Work

Any industry where knowledge is scattered across thousands of documents and multiple teams.

🏥

Healthcare

Clinical research, protocols, treatment literature

⚖️

Legal

Contracts, regulations, case precedents

🎬

Media & Broadcasting

Archives, metadata, rights, production

🔬

Research

Literature reviews, cross-disciplinary links

💻

Software

Codebase intelligence, architecture, docs

🏛️

Government

Air-gapped, sovereign data, compliance

"On a 193-page book: 90% verified precision on causal relationships spanning hundreds of pages."

— From the paper, Source verification study

On-Premises First. Zero Data Leakage.

Your documents never leave your infrastructure. Period.

Noēsis runs entirely on your hardware — a single workstation GPU is enough. No cloud API calls during ingestion or querying. Perfect for healthcare, defence, legal, and any environment where data sovereignty is non-negotiable.

Need cloud power? Plug in any LLM (OpenAI, Anthropic, AWS Bedrock). Same API, same results. Your choice, always.

12GB GPU

Minimum hardware. RTX 4080 Laptop class. Full system operational including 35B parameter model.

6GB Stress Tested

160+ minutes continuous operation. Zero crashes. The system adapts — it never fails catastrophically.

Multi-GPU Scalable

Start on one machine. Scale to multiple nodes. Same architecture, linear performance gains.

Backend Agnostic

Local GPU, Ollama, OpenAI, Claude, AWS Bedrock. Swap backends without changing anything else.

Works With Your Tools

Universal MCP protocol. Drop-in integrations for the tools your team already uses.

Claude Code

Anthropic's coding agent queries your knowledge graph natively.

Cursor IDE

Knowledge-aware coding without leaving your editor.

Gemini CLI

Google's CLI with graph traversal from the terminal.

Any MCP Agent

Universal adapter: works with Continue, Cline, Roo Code, and more.

Recent

Aug 2026

Paper published on arXiv

Full system paper with benchmark results, ablation studies, and architectural details.

Aug 2026

IAMT UK Summer Event — AWS HQ London

Presented "The Sector-Agnostic Shift" alongside James Whitebread (CARE ADHD).

Aug 2026

Italian Patent Filed

Application No. 102026000023146 — covering all four core algorithms.

Sep 2026

IBC Amsterdam

Coming: RAI Amsterdam, 11–14 September 2026.

See Noēsis in Action

Schedule a demo. See your own documents transformed into an interconnected knowledge graph in minutes.

Or visit us at IBC Amsterdam, 11–14 September 2026