Coelanox

coelanox — session

$ coelanox --about

Coelanox turns your trained model into a sealed, hardware-ready binary.

Not a script or a framework—just a .cnox container and a tiny runtime.

AI unikernel: verify → map → execute. No Python, no Docker, no OS in the hot path.

$

> read the spec

$ cat README — system overview

1. The Format .cnox

Single, cryptographically sealed .cnox file. Inside: a compact math graph (51 ops), compressed weights, precompiled kernels (SIMD/GPU), and a rigid execution plan.

2. The Packager

Offline compiler/linker. Ingests framework graphs, strips framework bloat, converts math into a tiny Universal IR, and stitches vendor kernels (CLF) into the .cnox payload.

3. The Runtime

Tiny, Turing-incomplete engine. No web server, no network stack. Given a .cnox and input floats it only does three things: Verify (SHA-256), Map (mmap, Harvard-style code/data separation), Execute—produce logits.

contact — early-access

$ contact --early-access

Request early access, partnerships, or technical details.

> contact

$ coelanox --help

Try the CLI in your browser. Type: coelanox --help, coelanox package --help, coelanox --version

coelanox 0.6.0-beta — try in browser
Try: coelanox --help, coelanox package --help, coelanox --version
Or: coelanox validate -f model.cnox, coelanox env, coelanox run -f model.cnox -o out.json
$

$ tail -f CHANGELOG

CHANGELOG
  • [May 2025] Website revamp; terminal UI; CLI try-it terminal; positioning update.
  • [Apr 2025] Core .cnox format and runtime pipeline in development.
  • [Mar 2025] Packager IR and compiler architecture finalized.