NETSERVE / OPEN LAB

NO ACCOUNT. NO INSTALL. NO DATA UPLOAD.

Small networks.
Real behavior.

Explore exact computation encoded in neural weights. These demos load the same compiled threshold graphs used by Netserve and evaluate them right in your browser.

Browser demos use a JavaScript network evaluator. Download Netserve to run native Goose, the full server and local STUN/TURN chat. This page is not an Internet TURN relay.

01 / LOSSLESS CODEC

Pack text into fewer bits.

Letters, digits, spaces and periods. Up to 32 characters. A fixed 6-bit alphabet with a length byte; not a trained universal compressor.

Ready to run the encoder and decoder.
02 / NEURAL STATE

A counter with memory.

The network computes the next state and the JSON response. State stays in this tab and wraps from 255 to 0.

000
State: 8 bits
03 / NEURAL ECHO

Bytes through a network.

Send up to 64 UTF-8 bytes through a compiled identity graph. This demonstrates exact byte preservation, not an AI conversation.

Ready.
04 / RUN THE FULL SERVER

Take the lab home.

Native Goose, compiled pages and handlers, learned associative memory, local relay chat, TLS and authentication. Node.js 24+ is required.

Setup instructions →

OPEN MEASUREMENT

Neural behavior coverage

Loading the documented inventory...

This is our equally weighted feature inventory, not an independent certification or a percentage of source code, CPU time or memory. Secure-host services, browser UI, build tools and historical experiments are excluded. Host protocol parsing and assembly are counted in the inventory.

Inspect every counted behavior

    Download the inventory JSON · Inspect the measurement script · Build and test evidence

    What “neural” means here

    These are deterministically compiled threshold networks: weights and connections encode the supported computation. They do not learn from your input. The full server keeps sockets, cryptography, authentication and resource limits in a conventional secure host.