README.md

CORTEX TEST KIT

One-download bundle for local testing of the Portable AI / CORTEX project.

Contents

• portable-ai-runner/ — the full package: index.html (browser app + LAN mesh UI), serve.py (launcher + LAN mesh hub, HTTP :8080 / WS relay :8081), setup.py, stack/ (docker-compose.yml + Dockerfile.runner: code-server on :8443, runner on :8080), models/ (cores.json with the MEASURED MoE tier: Ornith-35B 122.7 tok/s, Qwen3-Coder-30B 146.3 tok/s, Ornith-9B 76.7 tok/s), scripts/ (all CORTEX demos + knowledge-graph builder), docs/ (architecture, benchmarks, system prompt).
• pyggi/ — verified mirror of https://github.com/coinse/pyggi (MIT), the genetic improvement / automated-repair engine wired as symbol §EVO_AST:PYGGI (see portable-ai-runner/docs/EVO-SKILLS-BRIDGE.md).
• MANIFEST.json — upstream commit + sha256 of every mirrored pyggi file and of the docker-compose.yml, so any tampering is detectable.

How to test

1. Download: huggingface-cli download Johnny5b/cortex-test-kit --repo-type dataset --local-dir cortex-kit && cd cortex-kit
2. Docker route: cd portable-ai-runner/stack && docker compose up --build → code-server (VS Code in browser) at http://localhost:8443, runner at http://localhost:8080.
3. No-Docker route: cd portable-ai-runner && python serve.py → http://localhost:8080.
4. pyggi skill: cd pyggi and follow the upstream README (Python + PyYAML). Provenance: commit 14361bb825cd386e9a9e78b58b3975c195f8fc7d — verify with MANIFEST.json before trusting the mirror.

Built 2026-09-18 by the CORTEX session; model weights are NOT included here — they are fetched to the drive by the manifests'' download scripts (models/download_models.py).