VINE Suite

Management tools powered by geometric intelligence

VINE Suite wraps proven open-source business tools in a geometric reasoning engine that learns during operation. Every module shares one core. Every decision is traceable. When one module encounters something new, all modules benefit.

No cloud API calls. No retraining cycles. No black box.

Ernest (ERP)
Full enterprise resource planning built on ERPNext, enhanced with geometric intelligence. Supplier scoring, inventory, payments, manufacturing, shipping, asset depreciation — 27 modules patched, 180+ tests passing. Every hard-coded threshold replaced with continuous geometric reasoning. Your business logic doesn't get rewritten — it gets smarter.
Try the ERP Demo →
Data Cruncher
High-volume text classification powered by geometric reasoning. 177 million items per second on a single consumer GPU. She classifies, surfaces ambiguity honestly, and works through edge cases with you. No training step. No cloud dependency. Traceable decisions at every level.
Try the Data Cruncher →
Risk Assessment
Geometric risk classification across operational domains. The same engine that classifies text data evaluates supplier risk, flags quality anomalies, and detects threat patterns — all without rule-based thresholds. Continuous assessment that adapts to new patterns in real time.
See Risk Demo →
Garden Geometry
Geometric gardening intelligence for planting lifecycle analysis, harvest readiness, crop approval workflows, species confidence resolution, data gap triage, and seed tradability assessment. Six geometric functions replacing hard-coded rules with continuous geometric reasoning.
Try Garden Geometry →
Flavour Engine
Ingredient topology for pairing, substitution, and dietary filtering. 250+ ingredients mapped across 7 geometric taste axes — sweet, salty, sour, bitter, umami, fat, spice. The same basin settling that runs ERP decisions runs flavour combinations. Early toy — geometry works, recipes still growing.
Explore the Topology →
Barry
A vanilla GPT-2 running a vending machine. No finetuning, no text parsing — the model's logit entropy, hidden states, and probability spread directly drive stock management, pricing, and customer service through geometric settling. Three concerns compete: desire, freshness, and discount. The only demo where the model IS the product.
Meet Barry →
🏥 Triage Engine 📷 Vision & Import ♟️ Bot Chess

What Makes This Different

Glass Box

Every decision maps to a specific geometric position. You can trace the full path from input to output, inspect what settled where, and correct it at the point of interaction. When VINE pulls a sentence from memory, the chat intelligence panel shows you exactly which source she drew from, what her confidence levels were, and where the response sits in geometric space. This isn't explainable AI bolted on after the fact — traceability is the architecture.

Runtime Learning

VINE learns during operation. When the ERP encounters a supplier pattern that doesn't fit existing categories, she requests new dimensions at runtime. The vocabulary grows because the software is learning. No retraining, no fine-tuning, no parameter freeze.

One Core, Every Domain

The algorithm that runs your ERP is the same one that classifies 400,000 tweets, plays chess, generates terrain, and holds conversations. When data classification encounters a new pattern, that learning propagates across the entire system. Each module you add makes every other module smarter.

Drop-in Integration

VINE doesn't replace your codebase. She makes it intelligent. Patch your existing files, and every conditional becomes continuous geometric reasoning. Already proven on ERPNext — 27 modules, no rewrite.


Performance

Benchmarked on real data. Single RTX 4070 SUPER (~£600 hardware).

1K items → 5.2M/sec  •  0.19ms
10K items → 47.4M/sec  •  0.21ms
50K items → 152.5M/sec  •  0.33ms
61K items → 177.0M/sec  •  0.35ms

1.2MB VRAM. Sub-millisecond latency at every scale. The same algorithm runs on 1983 BBC BASIC hardware at 55,555 operations per second. The geometry is efficient — not just "fast because GPU."

For context: X/Twitter runs 20,000 GPUs for content classification. This does comparable work on one consumer gaming card.


What the Industry Is Spending Billions On

Real-Time Learning

The industry approach: fine-tune or RAG after the fact. 200-person teams building smart devices that can't learn your habits while running.

VINE learns during operation — a VINE-powered device would learn habits geometrically with zero overhead and no LLM calls.

Agent Coordination

The industry approach: multi-agent orchestration with 15× token cost and compounding errors. Each agent isolated behind its own context window.

VINE's geometric settling provides natural coordination — multiple concerns resolve to one stable state without orchestrators or message passing.

Compute Efficiency

The industry approach: scale to millions of GPUs. Memory famine warnings. $20 billion hardware investments.

VINE runs on a consumer laptop. Same algorithm, same results, fraction of the power.

Embodied Reasoning

The industry approach: massive vision-language-action models that treat robotics as a separate engineering problem.

VINE is already a persistent learning process. The same core that classifies data, generates terrain, and runs security can take sensor input. Adding motors is wiring, not architecture.


Ready to see it working?