01 · Local
Bring your own engine
Already run Hermes, Ollama, llama.cpp, LM Studio or vLLM? The harness scans the loopback ports, reads the model list, and opens a console. No model download required.
Engine compatibilityNo API key · loopback only · your hardware
Run your engine.
Use it in the browser.
LYGO Local is the harness that finds the AI engine already on your machine — Hermes, Ollama, llama.cpp, LM Studio, or vLLM — and opens a clean local console. No account, no cloud key, no prompt sent to excavationpro.ca.
Local console + public relay are live. The Windows setup is signed with the LYGO Local self-signed certificate. It is a preview certificate, not a public CA; Windows may show Unknown publisher until you trust it.
Three ways to use LYGO
The API portal stays for people with keys. LYGO Local is for people who want their own engine and their own hardware.
02 · Full pack
LYGO Studio Pack
A hardware-matched model pack from our Hugging Face vault. Text and code first; image, audio and video as separate packs because one 9B model cannot do all four well on 8 GB.
Model packs03 · API
Public API portal
No local engine, no install. Paste a Groq, Gemini, OpenRouter, OpenAI, Claude, Grok or DeepSeek key into the existing portal. Your key stays in the tab.
Open API portalDownload LYGO Local
Signed preview available. The setup is Authenticode-signed with the LYGO Local self-signed code-signing certificate. Windows may show Unknown publisher because this is not a public CA certificate. The public .cer, SHA-256 and trust notes are beside the download. No administrator rights are required.
Trust note. The signature is real but the certificate is self-signed. Verify the SHA-256 before running. If you trust this source, you may import LYGO-Local-CodeSigning.cer into your user Trusted Publishers store; otherwise use Windows “More info → Run anyway” only after comparing the hash. This is not a claim of public-CA trust or SmartScreen reputation.
What Setup does. It installs per-user into %LOCALAPPDATA%\LYGO, writes the harness, uninstaller, certificate and README, creates a Start Menu shortcut, and launches LYGO Connect. No service, driver, firewall rule, or startup entry. Uninstall from the installed LYGO-Local-Uninstall.exe; your models and engines are untouched.
- Signed setup + signed harness + signed uninstaller
- Per-user, no admin, no Windows service
- Loopback bind only, never LAN
- Engine boot only on local button press
- No prompt upload to excavationpro.ca in local mode
- Public certificate and SHA-256 published
- Uninstaller removes app files and shortcut only
- Python/Go/Node dependencies not required
Phase 2 relay is live. LYGO Connect can display a six-character pair code and connect to the Cloudflare relay; the pairing preview is at /portal/local/pair.html. Local-first API fallback is available on that page.
Developer source build
Requires Go 1.25+. This is for review; the signed setup is the recommended install.
cd lygo_local go build -trimpath -ldflags "-s -w" -o dist\lygo-connect.exe . .\dist\lygo-connect.exe
Engine compatibility
The harness only probes these loopback ports. It does not scan your LAN or the internet. Custom adapters live in a local config file the website cannot write.
| Engine | Default endpoint | Detected | Boot from harness |
|---|---|---|---|
| Ollama | 127.0.0.1:11434 | Yes | Yes, if ollama is on PATH |
| Hermes / llama.cpp / OpenAI-compatible | 127.0.0.1:8080, :8081 | Yes | Start via your launcher; custom config can add a boot command |
| LM Studio | 127.0.0.1:1234 | Yes | Start in LM Studio; harness connects |
| vLLM | 127.0.0.1:8000 | Yes | Start via your vLLM command |
| LYGO Console | Custom adapter | Yes | Add its launcher in %LOCALAPPDATA%\LYGO\engines.json |
The first build was tested against a live QwenPaw 9B server on :8080; the harness read qwenpaw-9b and returned a local chat answer. No cloud request was involved.
Model packs from Hugging Face
Models are separate from the harness. They are large, licensed, and hardware-specific. Pull only what your machine can run.
Text + code
LYGO Console model vault
GGUF models used by the LYGO console, including Qwen-family text and coding models. Start with the 9B Q4 tier on 8 GB.
Open model vaultLarge expert
LYGO Qwen 3.6 35B OpenClaw
MoE GGUF for bigger GPUs. Not a starter pack; expect a large download and serious RAM/VRAM requirements.
Open modelEngine + kits
LYGO console builds
Installers, console builds and supporting files. Verify every SHA-256. Model licences remain with their authors.
Open buildsCreative
LYGO Resonance Engine
Image-to-sound and creative profile engine, separate from the text LLM. Useful once the local console is running.
Open engineImage / audio / video
Separate media engines
Qwen-Image/SDXL, ACE-Step and Wan 2.2 are separate runtimes, not one magic model. The Studio Pack will detect hardware and recommend a tier.
All LYGO modelsAlready have one?
Bring your own
Ollama, LM Studio and llama.cpp already manage their own downloads. LYGO Connect will use them without duplicating a model.
See portsFull disclosure
What the harness does, what it does not do, and what leaves your machine.
What it reads
Your local loopback ports and the model list returned by the selected engine. Optionally, a local engines.json you create.
What it sends
Only your prompt, to the selected engine on 127.0.0.1. No prompt is sent to excavationpro.ca. No analytics endpoint is called by the harness.
What it stores
No chat history is written to disk. The browser tab holds the current answer. The harness logs startup and engine status to its own console.
What it will add later
Phase 2 pairing sends a one-time code and session token to the relay. That will be disclosed and opt-in. The relay cannot see prompts if WebRTC is used; with a WebSocket relay it can, so that choice will be explicit.
What it never does
No arbitrary command from the website, no LAN scan, no open proxy, no silent auto-start, no remote shell, no model upload.
Your responsibility
Model licences, hardware limits, content you generate, and any API or tip you choose. The harness is a local tool, not a hosted service.
Legal: LYGO Local terms & EULA · LYGO Local privacy notice · Portal legal · Site privacy · Site terms.
What Phase 1 already proves
This is not a mockup. The built harness was run on this machine against a live local engine.
Detected
QwenPaw 9B on 127.0.0.1:8080; model qwenpaw-9b; auth handled locally with the engine's key.
Answered
The local browser console sent a prompt through the harness and received OK. No cloud API key was used.
Bounded
Loopback only, token-gated local API, no CORS, no prompt upload, no engine boot without a local button press.