01 / NEUROX V6 · CURRENT LIVE MODEL · 897M PARAMETERS
Intelligence.
Inside your browser.
● V6 / CURRENT · LIVE INFERENCEV5 / IN DEVELOPMENT
Real neural inference on your own device. NEUROX CORE runs a 897M-parameter ONNX model locally via WebAssembly. This is an experimental proof of concept, not a production assistant. The 196 MB INT8 model is downloaded on demand.
NEUROX CORE / LIVE CONSOLENEUROX V6 not loaded
02 / MODEL ARCHITECTURE● DEVELOPMENT PIPELINE
PARAMETERS897MCompact autoregressive transformer
TRANSFORMER BLOCKS12Decoder-only architecture
CONTEXT WINDOW2,048Maximum supported token context
VOCABULARY16,384BPE tokenizer entries
INFERENCE / LOCAL EXECUTION LIVE PIPELINE VISUALIZATION
INPUT
TEXT
TEXT
BPE
TOKENIZER
TOKENIZER
ONNX
TRANSFORMER
TRANSFORMER
SAMPLING
OUTPUT
OUTPUT
Inference runs in the browser using ONNX Runtime Web / WebAssembly. Model weights are downloaded to the visitor's device; responses are generated locally. Browser caching depends on server headers and browser settings.
03 / V5 DEVELOPMENT ROADMAP
Training and development
NEUROX V6 is the current operational browser model. NEUROX V7 is in development.
01BASE MODELAVAILABLE
02BROWSER INFERENCEACTIVE
03CRYPTO DATA CURATIONPLANNED
04DOMAIN FINE-TUNINGPLANNED
NEUROX CORE / 897M MODEL · Experimental browser-native inference. Uses ONNX Runtime Web from jsDelivr CDN. INT8 browser compatibility is experimental. Prompts and inference stay on the visitor's device; CDN access and browser compatibility are required.