Yes. It is MIT licensed for you to install and use. YSK Limited does not charge for that use. Install, workers, or a custom build: see the Business tab — no prices on this page.

How is YSK Omni different from the Grok CLI Gateway?

The Grok CLI Gateway (gctoac) wraps the local Grok CLI. YSK Omni runs local Hugging Face weights: text through llama-server or vLLM, and media through separate workers.

What happens if a media worker is not configured?

Image, speech, and transcription requests return 501 engine_unconfigured until their workers are set. Without a video worker, the gateway may return a short H.264 sample rather than a generated video.

YSK product · MIT · npm · v1.0.2

YSK Omni

Local Hugging Face weights behind an OpenAI-compatible HTTP gateway. CLI: ysk-omni / ysko, default port 3850.

GitHub npm Business

ysk-omni / ysko
Port 3850
MIT

YSK Omni concept: hexagonal gateway node with a local host

OpenAI APIChat · Responses · Messages

AES-256-GCMChats stored encrypted in SQLite

OTP adminOne-time login · five minutes

Catalog LoadGGUF · llama-server · vLLM

Local weights → a live gateway

Keep the SDKs, curl, and Open WebUI you already use. The models stay on your machine. The gateway does not launch an external coding CLI.

ClientOpenAI SDK / curl / Open WebUI
Gateway :3850Auth · limits · queue · admin
Textllama-server · vLLM · echo
Media workersImage · speech · transcription · video

Client (OpenAI SDK / curl / Open WebUI)
        │  Authorization: Bearer omni_live_…
        ▼
   Express gateway :3850
   · auth · rate limit · safe/agent · queue · Admin
        │
        ├─ GGUF text      → llama-server
        ├─ safetensors    → vLLM
        ├─ echo           → built-in protocol test
        ├─ image          → OMNI_IMAGE_URL
        ├─ TTS / STT      → OMNI_TTS_URL / OMNI_STT_URL
        └─ video          → OMNI_VIDEO_URL (or a short H.264 sample)

Local weights

Hugging Face GGUF and safetensors live under ~/.ysk-omni/models/. Pull first, then Load before you chat.

OpenAI-compatible

Chat Completions, Responses, Anthropic Messages — existing SDKs can point at the local gateway.

Catalog Load

Hub search, download queue, Load / Unload. GGUF uses llama-server; safetensors use vLLM.

Encrypted chats

Conversations are stored with AES-256-GCM in SQLite. Admin opens with a one-time code, not a live key in the browser.

Feature	Detail
OpenAI API	`POST /v1/chat/completions`, `POST /v1/responses`, `GET /v1/models`
Anthropic API	`POST /v1/messages` (Bearer or `x-api-key`)
Media	Image generate / edit, video, `/v1/audio/speech`, `/v1/audio/transcriptions`
Catalog	Hub search, download queue, local models, Pull / Load / Unload / delete
Runtimes	Install / uninstall llama-server, ffmpeg, and so on per OS
Keys	`omni_live_…`; roles admin / client; modes safe / agent; allowed models
Encryption	AES-256-GCM chats under `~/.ysk-omni`
Admin	OTP login; dashboard, playground, catalog, media, queue, DDoS, PM2
Queue	Pause / drain, jobs, priority, dead-letter purge
DDoS	Rate limits, auto-ban, blacklist, reverse-proxy IP source
CLI	Lifecycle, keys, catalog, queue, abuse controls, doctor, self-update
Ops	SQLite, PM2, Node.js 20+, default port 3850

API and protocols

Endpoints, media workers, error codes, and how this differs from the Grok CLI Gateway.

Protocols and endpoints

Base http://127.0.0.1:3850; auth Authorization: Bearer omni_live_…
GET /health; GET /ready database + text engine (llama-server on PATH)
GET /v1/models lists ids this key may use, with modality and runtime
POST /v1/chat/completions (optional stream: true); POST /v1/responses text subset
POST /v1/messages Anthropic style, also accepts x-api-key
Gateway extras: omni.sessionId, omni.asset_ids. OpenAI SDKs ignore unknown fields
Documents and assistants: /v1/documents and related OpenAI-shaped routes (feature flags)

Media workers

POST /v1/images/generations, POST /v1/images/edits
POST /v1/videos, GET /v1/videos/:id, GET /v1/videos/:id/content
POST /v1/audio/speech, POST /v1/audio/transcriptions
Point OMNI_IMAGE_URL / OMNI_TTS_URL / OMNI_STT_URL / OMNI_VIDEO_URL at OpenAI-shaped workers
No image / TTS / STT worker → 501 engine_unconfigured
No video worker may return a short H.264 sample, not a generated video
Image / video / speech need an agent or admin key, unless the caller is an Admin OTP session
Output types: png / jpeg / mp3 / wav / mp4; ffmpeg can convert

Error codes

JSON: { "error": { "message", "type", "code" } }
unauthorized 401 · forbidden 403 · model_not_allowed 403
feature_disabled 403/501 · engine_unconfigured 501 · engine_timeout 504
rate_limit_exceeded 429 · queue_full 429 · queue_draining 503

Versus Grok CLI Gateway

gctoac wraps the local Grok CLI (grok -p), default port 3847
YSK Omni runs local Hugging Face weights: text via llama-server or vLLM, media via workers, default port 3850
The gateway does not launch an external coding CLI
Both are MIT, OTP admin, and OpenAI-compatible

Admin

Ops panel after OTP login. UI languages: English and Hong Kong written Chinese. URL http://127.0.0.1:3850/admin/.

Login

ysk-omni admin otp: five minutes, single use
The panel does not store a live API key in the browser
ysk-omni admin off disables the panel; only admin on turns it back on

Catalog and runtimes

Pull does not start until you press Pull on that row, or run ysk-omni pull.

Hub and local models

The Hugging Face tab uses the official Hub REST API; sync caches the top 50 GGUF ids by downloads (metadata only)
Rows without a local runtime are hidden; size is MB or GB, with estimated VRAM when known
In-flight pulls show percent, size, speed, and ETA; leaving the tab does not drop that block
Pulled GGUF and snapshots (whisper, diffusion, T2V) appear as local models: Load, Unload, delete

After Load

Loading GGUF starts llama-server for persistent chat
Safetensors text ids need vLLM
Diffusion / whisper / T2V are used from disk by media workers, not llama-server
Built-ins include echo, piper/lessac-high, tts-1, whisper-1

Runtimes

ysk-omni runtimes: host OS, installed / missing, install plans
runtimes install <id>: Homebrew / pip / winget / Docker (llamacpp, ffmpeg, mlx, …)
Admin uninstall asks for confirmation first

Queue and security

Encryption, keys, queue, DDoS, and operational notes.

Encryption and data directory

Chats are AES-256-GCM in ~/.ysk-omni/data/gateway.db
Weights under models/; .env holds ENCRYPTION_KEY, port, worker URLs
The plaintext key is shown once at create; lists never include the secret

API keys

Format omni_live_…; roles admin / client; modes safe (public) / agent (full media and tools)
Per-minute rate, IP / CIDR allow-list, allowed-model list
Empty model list = unrestricted; not on the list → 403 model_not_allowed
Admin OTP sessions are not model-restricted
CLI: ysk-omni key create|list|show|update|revoke|activate

Chat queue

Stats, policy, pause / resume, drain / undrain
Jobs, cancel, requeue, priority, dead-letter purge
Full queue 429; draining 503
Admin and CLI (ysk-omni queue …) share the same operations

DDoS / abuse

Limits per key / IP, chat burst
Auto-ban: failed auth, repeated 429, concurrency flood
Presets relaxed / balanced / strict; ddos ban / unban / blacklist
Reverse-proxy hops and IP source (auto / Cloudflare / nginx, …)

Operational notes

Back up ENCRYPTION_KEY — without it, history cannot be decrypted
Use safe-mode keys on the public internet; keep admin on localhost or a VPN
OTP is five minutes, once; do not paste a live key into the browser
Do not commit .env; do not leak OTP or admin keys

CLI and operations

ysk-omni / ysko (not yo). Global flags: --home, --port, --json.

CLI

Lifecycle: setup, start (-f / --pm2), stop, restart, status, doctor, migrate, seed, open, version, update
Admin: status / on / off, otp, sessions list / revoke
Keys: create (-r -m --models), list, show, update, revoke, activate
Catalog: catalog, search, sync, show, pull, models, load, unload, rm
Runtimes: runtimes, install, uninstall
Queue / DDoS: stats, policy, pause / drain, jobs, ban / unban
Observe: docs, chats, conversations, audit, stats, logs
API features: api features get|set; presets open / locked / dev

Install, port, production

Node.js 20 or newer
Default port 3850; admin /admin/; API /v1
Data directory ~/.ysk-omni (--home)
ysk-omni doctor: Node, env, loaded models, VRAM, build, runtimes, port clashes
Production: start --pm2; update --check for self-update

Concept art and a quick install

These are product concept stills, not admin screenshots. Open /admin/ after install for the real UI.

Catalog

Hub search, download queue, local models. Concept art.

Load engine

GGUF via llama-server; safetensors via vLLM. Concept art.

Admin

One-time login, not a live key. Concept art.

Media workers

Image, speech, transcription, video on separate paths. Concept art.

Queue

Pause, drain, priority, dead letters. Concept art.

Local host

Weights and gateway stay on your machine. Concept art.

Quick commands

Install globally
Setup, then start
Admin one-time code → open /admin/
Catalog pull → Load → call the API

npm install -g ysk-omni
ysk-omni setup
ysk-omni admin otp
ysk-omni start
# http://127.0.0.1:3850/admin/

Need install, media workers, or a custom build? See Business — no prices on this page.

Open source to use. YSK can install or custom-build.

YSK Omni is MIT-licensed. The gateway and weights sit on your machine. YSK Limited can install it, attach media workers, or custom-build on ysk-omni — no fee for open-source use; custom work is quoted in writing.

Use it yourself

For self-hosting, internal trials, and open-source work.

Local Hugging Face weights + OpenAI-compatible gateway
npm install; port 3850
No charge for that use

YSK install / custom

For production, workers, white-label, or a private network.

The host stays yours — not a multi-tenant cloud gateway
Hong Kong company contract; NDA when needed
No custom prices on this page; quotes are written

Local / on-prem language models

Weights stay on disk. Point your existing OpenAI SDK at the gateway.

You already have OpenAI-shaped clients

curl, SDKs, Open WebUI → private GGUF / safetensors.

Media pipelines

Wire image, speech, transcription, and video workers to OMNI_*_URL.

Harden before go-live

Key policy, DDoS, admin on localhost only — executed to a written scope.

Custom scope	Detail
Install	Node.js 20+, `ysk-omni setup`, OTP, first catalog Pull / Load, PM2
Media workers	`OMNI_IMAGE_URL` / TTS / STT / VIDEO; ffmpeg; unset routes still 501 or a short sample
White-label	Admin name, logo, default locale
Keys and defence	Role / mode, allowed models, rate, IP allow-list, DDoS presets
Integrations	Existing OpenAI-shaped clients, gctoac, internal accounts, CLI / agent scripts
Private network	Admin on localhost or VPN; weights stay on your disk
Runtimes	llama-server, vLLM, ffmpeg install plans per OS
After delivery	Docs, remote training, warranty window. Small changes can use the outsourcing monthly tech tickets

How we work

Every step has a written scope. Open-source download and install / custom are billed separately.

EnquireWhatsApp or email: host, VRAM, models and workers
DiscoverRemote call: OS, network, compliance, white-label / data residency
ProposeScope, timeline, deliverables (install vs custom modules); written quote
ContractHong Kong company; NDA when needed
BuildOn ysk-omni; staged demos
AcceptGateway responds, /admin/ logs in, docs, training
WarrantyFixes in the agreed window; then maintenance or outsourcing monthly

Deliverables (per contract)

A working gateway and admin on your host
Deploy and ops docs; remote training
Defect fixes in the agreed window
Open-source core stays MIT; custom modules follow the contract

Honest limits: YSK Omni is not multi-tenant SaaS. YSK does not turn your GPU into a YSK cloud. Unconfigured media workers still return 501 or a short H.264 sample — custom work does not rebrand that sample as a full video engine. We do not apply for Hugging Face or cloud accounts on your behalf.

Business questions

Is the open-source version free?

Yes. MIT licensed. YSK does not charge for that use. Install, workers, and custom builds are quoted by scope.

Why are there no prices here?

A single-host install and a worker + white-label + private-network build differ a lot. Quotes are written. Small changes can use the outsourcing monthly tech tickets; large work is a project.

Does custom work become a YSK-hosted cloud gateway?

No. It stays a local gateway. You own the host and the weights, or the contract names them. YSK does not run a multi-tenant SaaS edition of YSK Omni.

WhatsApp for custom work [email protected]

FAQ

Is YSK Omni free and open source?

Direct answer: Yes. MIT licensed. Install, workers, or a custom build: see Business — no prices on this page.

How is it different from the Grok CLI Gateway?

Direct answer: gctoac wraps the local Grok CLI. YSK Omni runs local Hugging Face weights through llama-server or vLLM.

What are the default port and CLI?

Direct answer: Port 3850. The CLI is ysk-omni, short name ysko. Admin opens with a one-time code.

What if a media worker is not set?

Direct answer: Image, speech, and transcription return 501 engine_unconfigured. Without a video worker you may only get a short H.264 sample, not a generated video.

How are chats stored?

Direct answer: AES-256-GCM in SQLite under ~/.ysk-omni. Back up ENCRYPTION_KEY.

Which Node version is required?

Direct answer: Node.js 20 or newer. Install ysk-omni globally with npm.

Start with YSK Omni

MIT licensed · YSK Limited. Point the clients you already use at local Hugging Face weights. Install or custom work is quoted in writing.

Source npm WhatsApp

YSK Omni

Local weights

OpenAI-compatible

Catalog Load

Encrypted chats

Protocols and endpoints

Media workers

Error codes

Versus Grok CLI Gateway

Login

Pages

Hub and local models

After Load

Runtimes

Encryption and data directory

API keys

Chat queue

DDoS / abuse

Operational notes

CLI

Install, port, production

Catalog

Load engine

Admin

Media workers

Queue

Local host

Quick commands

Use it yourself

YSK install / custom

Local / on-prem language models

You already have OpenAI-shaped clients

Media pipelines

Harden before go-live

Deliverables (per contract)

Business questions

FAQ

Start with YSK Omni