lygo-console-builds

LYGO LLM Console — V1 installers (build 1.1.1)

Offline, local-first LLM console for Windows. One console, three systems:

Installers for USB LOCAL and PC LOCAL are here. The web portal is not installed; it is a page.

Get the version this repo is for: build 1.1.1, build-line 2026-09-20, signature Δ9Φ963-LYGO-LLM-CONSOLE-v1. Sealed to the canon vault (2026-09-20_build-1.1.1), clean-room restored and re-verified before release: 340 files / 50 folders on PC, 361 / 49 on USB, 605 tests passing, certify_build green, verify_install 26/26.

---

Which one do I want?

Light = the complete console, no model. It ships a fetcher and a pinned manifest, so you pull only the models you want, from our own hosts, hash-checked. Full = the same console with gemma4-12b (7,381,382,048 B) and its vision projector already inside. It installs and runs with no network at all.

The model that ships is Gemma 4 12B Unified — text, image and audio, Apache-2.0, legally redistributable. It boots and reads images on an 8 GB-VRAM box (it fits a 31.8 GB / RTX 4060 Ti rig comfortably) and scales up to bigger weights on better hardware without reinstalling the console.

The full installer is a 3-file set. …FULL.exe is disc 1; the two .bin files are discs 2 and 3. Keep all three in the same folder. Windows will not build a single .exe larger than ~4.2 GB, which is why it is split — this is normal Inno Setup disk spanning, not a broken download.

---

Verify first (30 seconds)

CertUtil -hashfile LYGO_LLM_CONSOLE_V1_PC_SETUP.exe SHA256

Compare against SHA256SUMS.txt — every character. If a digest does not match, stop and delete the file; do not run it.

Install

1. Run the installer (per-user, no admin prompt needed for the light builds).
2. Pick your folder. PC builds default to D:\; the USB build targets the stick.
3. Light: after install, open models\ and run the fetcher to pull models:

   python fetch_models.py --list
   python fetch_models.py --profile basic --dest <your models dir>
   

   It reads models.lock.json, fetches only from our hosts, and verifies every SHA-256 before the file is kept. --profile basic ≈ 7.3 GB, full ≈ 11.7 GB.
4. Launch LYGO_LLM_CONSOLE.bat, press Scan drives, boot the model.
5. Read READ_DISCLAIMER_FIRST.md in the install folder.

Silent install (scripted/offline): LYGO_LLM_CONSOLE_V1_PC_SETUP.exe /VERYSILENT /SUPPRESSMSGBOXES /NORESTART /DIR="D:\LYGO_LLM_CONSOLE".

---

Models and licences

Model weights do not live in this repo — they live in the companion repo lygo-console-models (GitHub mirror: https://github.com/DeepSeekOracle/lygo-console-models), so the light installers can fetch them from one place we control:

models.lock.json pins an immutable revision plus a SHA-256 per file. Nothing is fetched from a third-party site, and nothing is fetched from main (which could change under you). Engine: llama.cpp b10988. Fonts: Cormorant Garamond, IBM Plex Mono, Syne (OFL).

The console source itself is under the LYGO Sovereign License v3.0 — source available, free to use and build on; not MIT, not for resale, rebranding or white-labelling. Full terms and the legal summary: https://chatagent.ca/portal/legal.html

Trademarks belong to their owners. Not affiliated with or endorsed by Google, Alibaba, Meta, Anthropic, OpenAI, xAI, Mistral, or the llama.cpp project.

Support

Free downloads. Optional tip, never a paywall: https://www.paypal.com/paypalme/ExcavationPro

Steward: Justin Helmer (Lightfather / Excavationpro) · All builds are listed at https://chatagent.ca/lygoskillhub.html · Signature Δ9Φ963