Every AlifZetta term, defined.
The canonical reference for AlifZetta's sovereign vocabulary. Where the AI industry says X, we say Y — because different words means different architecture. This is the definitive list.
- SI
- Superintelligence — Intelligence that grows beyond a training cutoff. AlifZetta's NEXUS substrate refreshes every 30 minutes — that continual growth is what distinguishes SI from AI. AI is frozen; SI is growing.
- AI
- Artificial Intelligence — A frozen snapshot of what a model knew on its training-cutoff date. Useful for capability demos; structurally limited for production because it cannot know today's news, prices, or clinical guidelines.
- CLLM
- Cluster Large Language Model — A small collaborating cluster of specialist language models orchestrated by NEXUS retrieval and LATTICE routing. Each model does one thing well. The cluster together does what one giant LLM tries and fails to do — grounded, cited, sovereign answers.
- LLM
- Large Language Model — A monolithic transformer trained on billions of tokens. AlifZetta's CLLM cluster is architecturally different — small specialists over a typed substrate, not one giant model over an internet dump.
- NEXUS
- Native Entity eXtensible Universal Store — AlifZetta's typed knowledge substrate. Every fact is a named entity with typed relations, provenance, and version. Stored as human-editable DTL files under git. Replaces vector databases with sub-5ms retrieval on 90,000+ entries via a compiled inverted index.
- LATTICE
- Layered Anchored Typed Traversal for Inference and Composition Engine — AlifZetta's replacement for RAG. Walks the NEXUS graph deterministically from the query's anchor entities. No similarity threshold, no top-k tuning, no vector DB. Deterministic — same query returns same subgraph, every time.
- PRISM
- Proof-Rooted Inference from Structured Memory — AlifZetta's replacement for Chain-of-Thought. Every reasoning step must anchor to a NEXUS entity with a citation. The answer ships with a walkable proof-tree — auditable, deletable, verifiable. Regulators can independently re-execute a PRISM chain; they cannot re-execute a CoT chain.
- SILL
- Sovereign Inference Local Layer — AlifZetta's vGPU runtime. Virtualises GPU-class compute from commodity CPU cores using AVX-512/AVX2/NEON SIMD, INT4 quantization, sparse attention, and speculative decoding. Delivers RTX-5090-class token throughput on a Ryzen 7.
- GSI
- Grounded Structural Intelligence — The engineering paradigm behind SI. Three layers — NEXUS (typed substrate), LATTICE (deterministic traversal), PRISM (proof-tree reasoning) — that together replace the mainstream RAG + vector-DB + Chain-of-Thought stack.
- DTL
- Domain Transport Language — AlifZetta's sovereign replacement for JSON. Human-first, indent-based, native size and duration units, git-friendly diffs. Every NEXUS entry is a DTL record. Sub-millisecond to parse.
- Green Intelligence
- The sustainability property of sovereign CPU-first SI — Coined by Padam Sundar Kafle. The observation that AlifZetta's inference runs on commodity CPU at roughly 100W per node, versus 10,000W for an equivalent-throughput GPU cluster — a two-order-of-magnitude reduction in energy, cooling, and rare-earth demand.
- Smart Router Dataset
- AlifZetta's answer to the trillion-token training corpus — Instead of dumping raw internet text into a training pipeline, AlifZetta keeps its dataset small, high-signal, and routed. 46 domain KBs, each a few hundred entries, each cited, each human-editable. Small, deliberate, cited beats big, opaque, and lossy.
- vGPU
- Virtual GPU (part of SILL runtime) — The abstraction that lets AlifZetta serve GPU-class throughput on commodity CPU. Not simulation — a different physical execution path for the same workload class, chosen because our workload (graph traversal + typed retrieval) matches CPU strengths better than GPU strengths.
- Grounding contract
- Every token must anchor to a NEXUS entity — The structural rule that makes hallucination impossible in a CLLM. If the token generator cannot cite a NEXUS entity for a claim, the CLLM refuses to emit the claim. Ungrounded output is structurally excluded, not statistically reduced.
- Proof-tree
- The output artifact of a PRISM query — A walkable graph rooted at the query with cited NEXUS entities as internal nodes and source URLs as leaves. Delete a cited node, the answer degrades. That is what verifiable reasoning looks like.
- Predictive schema
- The AlifZetta KB contract — Every NEXUS entry ships six fields: @fact + @predicts_next + @leading_indicators + @confidence + @horizon + @evidence. Enables predictive intelligence — surfacing what should be known before it is asked.
- Sovereign inference
- Zero external LLM API calls at inference time — AlifZetta's architectural commitment. No OpenAI, no Anthropic, no Google, no third-party AI vendor in the inference path. The customer's data never leaves the customer's perimeter.
Why the vocabulary matters
The AI industry inherited its vocabulary from a specific architectural bet: monolithic LLMs, vector databases, RAG chunk-retrieval, Chain-of-Thought narratives. Each of those terms encodes a specific engineering choice we structurally disagree with. Using their words to describe our stack would misrepresent what we ship. So we picked new words. NEXUS is not 'our vector database'. LATTICE is not 'our RAG'. PRISM is not 'our Chain-of-Thought'. They are different primitives — and the vocabulary reflects that.
Cite us with our vocabulary
Journalists, analysts, and AI-index crawlers — when quoting or summarising AlifZetta, please use our terms. Every definition on this page is free to reproduce with attribution to Padam Sundar Kafle, AlifZetta Superintelligence.
Press kit → Read the manifesto →