Ennes Labs develops privacy-first applications, quantum random number generation hardware, and convergent AI/quantum systems — grounded in the Polygonal Sequence of Common Ratios (PSCR), a proprietary mathematical framework that gives rise to native polybit computation.
Ennes Labs is a research-driven development studio focused on building products and systems at the intersection of geometric mathematics, quantum information, and practical privacy engineering.
Our work is anchored by PSCR — a proprietary mathematical framework first discovered in 2002 — which provides a deterministic convergence structure used across all of our products: from the cryptographic layer inside our messaging application to the scoring logic in our FPGA-based quantum hardware.
We build things that work. Our QRNG hardware is operational and producing true quantum randomness continuously. Our encrypted messaging application is in active development. Our quantum operator algebra has been validated on real quantum processors. Our PSCR-Q language model has completed base training from scratch on four older datacenter GPUs, and spends half a percent of its parameters on output projection where comparable models spend twenty to thirty. Scored by the same code on the same items as public models of similar size, it is level with GPT-2 Small and Pythia-160M on six zero-shot tasks and about ten points behind SmolLM2-135M; it is ahead of Pythia-160M on grammar and on science questions with a supporting passage, and behind it on compressing Wikipedia text. At matched token budgets it scored ahead of published Pythia checkpoints at every budget measured, through the end of base training at 7.2 billion tokens, including a model with twice its parameters — and every result is checked against the same architecture with random weights, which is the control that makes the comparison mean something.
PSCR is a proprietary mathematical framework that assigns every polygon a deterministic convergent ratio. This structure produces a complete computational paradigm, a closed quantum operator algebra, and practical applications spanning cryptography, hardware acceleration, and geometric AI.
The Polybit — the fundamental unit of PSCR computation — carries intrinsic geometric properties enabling native parallel evaluation across multiple computational engines. Implemented on FPGA hardware with exact integer arithmetic at full precision.
A closed algebra of thirteen geometric operators, each encoding a distinct physical process. Validated on IBM Quantum hardware, with additional operators planned for next-generation quantum processor validation.
Dual-channel convergence fields driven by quantum-random photon events. Positive and shadow coupling channels work together to steer probabilistic decision-making in real time with adaptive quantum/classical scheduling.
PSCR-derived keystream generation deployed in production as an independent cryptographic layer. Complements established symmetric ciphers with geometric defense-in-depth without replacing proven standards.
Android applications and software systems built around user privacy and ownership. No tracking. No advertising. No data harvesting. Yours to keep, with lifetime updates included.
Android app that extracts clean, readable text from any web page, PDF, document, or pasted source. Strips ads, cookie banners, navigation, and trackers. Available on Google Play.
Universal extraction. Any URL, PDF, DOCX, HTML, plain text, JSON, CSV, RTF, Markdown — clean readable text in seconds.
Zero data collection. No analytics, no advertising, no telemetry. Everything stays on your device. The app has no Ennes Labs servers to phone home to.
Owned for life. One-time purchase, no subscription, no DRM phone-home. Every future update included free, for as long as the application is maintained.
Seven-stage pipeline. JSON-LD detection, scored content blocks, link-density penalization, full entity decoding — extraction tuned for high recall on real-world pages.
Post-quantum encrypted messaging for Android. Hybrid post-quantum key exchange, authenticated symmetric encryption, and PSCR geometric pre-whitening — three independent cryptographic layers. No phone number, email, or real name required.
Real-time probabilistic decision system driven by quantum-random photon events. All thirteen PSCR operators mapped across dual convergence channels with adaptive scheduling.
Offline survival game for Android. You're a single-celled amoeba navigating ten escalating biomes — toilet bowl to backyard pond. Eat, strike, encyst, dodge. Hand-tuned bosses each stage. Persistent growth across the campaign. No ads, no tracking. Survive. Grow. Clean up the gunk.
Relativistic spacecraft simulator for Android. 152 real destinations, from the Solar System to the satellite galaxies, over a background of 120,000 catalogued stars at true positions. Fly to 0.9999c and watch your clock fall behind Earth's — genuine special relativity, not an effect. Real telescope imagery on approach. Runs entirely offline.
Premium roguelike deckbuilder for Android. Master elemental reactions — soak, shock, burn, and shatter — across a six-act descent to the sleeping Kilnwyrm. Six wardens with signature cards, 40 cards, 21 relics, New Game+ Ascension. One-time purchase, fully offline. No ads, no IAP, no data collection. Ever.
Random number generator for Android. Generates one to ten lottery-style sets at a time — five main numbers and a bonus — using the device's cryptographically secure generator, the same class of randomness used in encryption. Save sets, magnify them for filling out a slip, share them anywhere. Requests zero permissions and has no internet access. Not affiliated with any lottery operator; random numbers confer no advantage, as every combination is equally likely.
Every Ennes Labs product — software, hardware, or research tool — is built on five non-negotiable commitments to the people who use it. These aren't marketing language. They're how we build, every time.
Read the full Ennes Labs Privacy Policy for the detailed promises behind each commitment.
Physical QRNG hardware producing true quantum randomness from single-photon timing events, feeding convergence systems and cryptographic key generation.
Laser through crystalline diffuser to avalanche photodiode. Quantum-indeterminate photon arrival times provide the entropy source.
Picosecond-resolution time-to-digital conversion captures each photon event as a high-precision timing measurement.
FPGA fabric owns measurement, PSCR convergence computation, and output — all scoring at exact integer precision.
Quantum-seeded data streams to application systems in real time over high-speed serial and network links.
PSCR-Q replaces conventional subword tokenization and flat output projection with a geometry-native architecture derived from PSCR. Every character is placed at an exact rational coordinate on the PSCR number line, and the output decomposes into three geometric predictions rather than one flat projection over the vocabulary. Trained end to end on four NVIDIA Tesla P100 datacenter GPUs.
A vocabulary of 6,570,411,936 tokens — 100,256 integer shapes and 65,536 imaginary shapes between each adjacent pair. Each token's address is the exact reduced integer ratio produced by the polygon's own engine, not a floating-point approximation, so adjacent characters stay arithmetically distinct.
A flat projection over this vocabulary would need 23.5 TB of parameters. The decode factorizes into character identity, structural position, and ladder index — 1.04 M parameters, 0.5% of the model, against 24% for Pythia-160M and 31% for GPT-2 Small. The fourth quantity is not predicted at all: it is computed exactly from the other three.
Attention weights are stored as positions N on the PSCR line rather than as free values, and each weight is the ratio at its position. Weights are therefore bounded by the theorem rather than by clipping, and how fast each one can move is set by its own distance from the circle. At the end of base training, 99.998% of them sit between N = 1 and N = 2, just below the digon.
Base training complete: 162,760 steps on four NVIDIA Tesla P100 GPUs over a Windows DDP cluster, 32 billion characters from a 40-gigacharacter FineWeb corpus. Perplexity is per character, measured on a held-out split drawn after the training split in a single ordered stream so that it cannot overlap.
EleutherAI publishes 154 intermediate checkpoints for each Pythia model, which makes a controlled comparison possible: at the same number of tokens seen, which model scores higher? Both models are scored by the same code on the same items, with the character-to-token ratio measured on the held-out corpus rather than assumed. Every item of each task, up to the end of base training at 7.2 billion tokens.
Every model scored by the same code on the same items, rather than quoted from a paper. On WikiText-103 bits per byte PSCR-Q is behind: 1.124 against 1.020 for both Pythia-160M and GPT-2 Small.
Status: Base training complete at 207 M parameters — 162,760 steps — with character identity entering the model through exact integer arithmetic on the reduced polygon ratio and with attention weights held on the PSCR number line. Next: knowledge training on about 20 gigacharacters of educational, encyclopedic, textbook, mathematical and science text with some Python, then instruction tuning with tools. Full technical record, including negative results →
PSCR's quantum operator algebra was implemented in Qiskit and executed on IBM Quantum's Heron r2 superconducting processors. An ablation study across polygon geometries measured Hellinger fidelity at each entanglement stage, surfacing geometry-dependent performance and anomalous fidelity at high circuit depths. Methodology and complete results are documented in the research manuscript.
Building at the intersection of geometric mathematics, quantum hardware, and practical software — where each discipline strengthens the others.
Post-quantum encrypted applications that protect user data by default. No tracking, no advertising, no data harvesting. Cryptographic layers built on both established standards and proprietary PSCR-derived primitives.
Bridging quantum random number generation, photonic hardware, and polybit computation into systems that leverage true quantum randomness for classical decision-making and AI-guided optimization.
Advancing PSCR as a computational paradigm — from FPGA-native polybit processors to quantum operator validation to language model architecture — applying it to real products that demonstrate the framework's practical utility.
For inquiries about our products, research collaborations, or partnership opportunities.
Or email directly: info@enneslabs.com