Muse Gateway

Self-hosted Muse session pool + OpenAI-compatible gateway (/v1/chat/completions, /v1/responses, /v1/tasks, /v1/models) with a web dashboard and Windows system-tray integration. It runs on your own machine with your own Muse accounts, and any tool that speaks the OpenAI API can use it.

License required. Anyone can install and open the dashboard, but the /v1/* gateway API stays locked (HTTP 402) until you activate a purchased license key. Buy a key at https://www.storetiki.online (floating or HWID-bound, time-limited) — support: Telegram @storetiki_bot.

Install

npm install -g muse-gateway

Run

muse-gateway

If muse-gateway is not recognized after installing, npm's global bin directory is probably not on your PATH — npx muse-gateway works in that case (npx resolves globally installed packages itself). To fix it permanently, add npm's global bin dir (usually %APPDATA%\npm, check with npm prefix -g) to your PATH.

Opens an interactive menu:

=======================================================
  Muse Gateway  v1.0.2
  Server:  http://127.0.0.1:1412   ● running — pro
=======================================================
> Open Dashboard (Web UI)
  Hide to Tray (Background)
  Run in this terminal (foreground)
  Enter License Key…
  Stop Server
  Exit

• Open Dashboard — starts the server if needed, opens http://127.0.0.1:1412
• Hide to Tray — runs in background with a system-tray icon (Open Dashboard / Auto-start / Quit) and a notification confirming you can close the terminal
• Run in this terminal — foreground server, Ctrl+C to stop
• Enter License Key — paste a MUSE1.… key, activates without opening the dashboard

Activate a license

Three ways — all equivalent:

1. muse-gateway menu → Enter License Key… → paste the key
2. Dashboard → Bản quyền tab → paste key → Kích hoạt
3. curl -X POST http://127.0.0.1:1412/api/license -H "content-type: application/json" -d '{"action":"activate","licenseKey":"MUSE1.…"}'

Until a valid key is activated, every /v1/* request returns 402 license_required. The dashboard and /api/* management routes always remain available.

Connect your tool

After activating and adding a Muse account in the dashboard, use these three values in any tool with a custom OpenAI endpoint:

from openai import OpenAI

client = OpenAI(base_url="http://127.0.0.1:1412/v1", api_key="YOUR_KEY")
reply = client.chat.completions.create(
    model="muse-chat",
    messages=[{"role": "user", "content": "Hello"}],
)
print(reply.choices[0].message.content)

Models

The model decides what you get back. The gateway never guesses from the prompt.

GET /v1/models lists every model with a kind field (chat, image or video). Names such as gpt-4o or claude-3-5-sonnet are aliases for clients that only offer those names; Muse answers them as muse-chat.

For a tool on another machine, turn on the tunnel in the dashboard and use the public address it gives you, such as https://u-xxxx.storetiki.online/v1. Through a tunnel an API key is always required, so treat that key like a password.

Generate images and video

Images: POST /v1/images/generations

The standard OpenAI image endpoint, so tools with a built-in image feature work directly:

curl http://127.0.0.1:1412/v1/images/generations \
  -H "Authorization: Bearer YOUR_KEY" -H "Content-Type: application/json" \
  -d '{"model": "muse-image", "prompt": "a cat on the grass", "size": "1024x1792"}'

size is mapped to the nearest ratio Muse understands (1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3). n returns up to 4 of the images Muse made. response_format is url (default) or b64_json. The response is { "created": ..., "data": [{ "url": "...", "revised_prompt": "..." }] }.

Video and long jobs: the task API

Video takes minutes, so use the task API: create a task, then poll it or receive a webhook. Results are stored on your machine.

Create a task

curl http://127.0.0.1:1412/v1/tasks \
  -H "content-type: application/json" \
  -H "authorization: Bearer YOUR_KEY" \
  -d '{
    "model": "muse-video",
    "prompt": "A cat playing piano on a rooftop at night",
    "duration": 30,
    "ratio": "9:16",
    "images": ["https://example.com/cat.jpg"],
    "webhook_url": "https://example.com/hooks/muse",
    "webhook_secret": "choose-a-secret"
  }'

The gateway answers 202 right away with a task whose status is running.

A URL that can't be loaded fails the request with 400, so a job never runs without the reference you sent.

Read the result

curl http://127.0.0.1:1412/v1/tasks/TASK_ID -H "authorization: Bearer YOUR_KEY"

{
  "id": "task_3f9c...",
  "object": "task",
  "status": "succeeded",
  "model": "muse-video",
  "requested_duration_sec": 30,
  "url": "http://127.0.0.1:1412/api/media/media-abc.mp4",
  "duration_sec": 30.1,
  "media": [{ "url": "http://127.0.0.1:1412/api/media/media-abc.mp4", "kind": "video", "duration_sec": 30.1 }],
  "partial": false,
  "text": "Your video is ready.",
  "thread_url": "https://muse.ai/thread/...",
  "account_id": "...",
  "error": null,
  "webhook_status": "delivered"
}

• status is running, succeeded, failed or cancelled. Poll every 10 seconds or so.
• url is the final file. media lists everything delivered.
• partial: true means Muse stopped before the requested length, and url is the longest clip it delivered. Open thread_url to check it.
• Media links are permanent only while the gateway keeps the file. Copy results you want to keep.

Other calls:

Restarting the gateway fails tasks that were still running; finished results stay readable. The last 500 finished tasks are kept.

Webhook

When a task ends, the gateway sends a POST to webhook_url, retrying twice if your server doesn't answer with 2xx:

{ "event": "task.succeeded", "task": { "id": "task_3f9c...", "status": "succeeded", "url": "..." } }

The event is task.succeeded, task.failed or task.cancelled. With a webhook_secret, verify X-Muse-Signature, which is sha256= plus the HMAC-SHA256 of the raw body:

import crypto from "node:crypto";

const expected = "sha256=" + crypto.createHmac("sha256", secret).update(rawBody).digest("hex");
const valid =
  signature.length === expected.length &&
  crypto.timingSafeEqual(Buffer.from(signature), Buffer.from(expected));

A webhook needs an address the gateway can reach. Calls that come through a tunnel can't target private network addresses, so localhost webhooks only work for local callers.

Use chat completions instead

POST /v1/chat/completions with muse-image or muse-video (or an image/video combo) also generates media, which suits tools that only speak the OpenAI API. Send stream: true, because proxies cut long silent requests. Extra fields are duration, thread, account, new_chat and idempotency_key. Images go in image_url parts, as URLs or data URLs, on the last user message.

Links appear as text under Attachments:, and the final chunk also carries them in structured form:

{ "muse": { "media": [{ "url": "...", "kind": "video", "duration_sec": 30.1 }], "partial": false, "thread_url": "...", "account_id": "..." } }

Good to know

• One job delivers up to 60 seconds. Split longer videos into several tasks.
• A long video can take many minutes. The gateway waits for the requested length, and returns what it has after 5 minutes without progress. Change this with MUSE_LONG_RENDER_GRACE_MS.
• Each account runs a few jobs at once; extra jobs wait in line.
• Audio input isn't supported.
• /v1/responses doesn't accept reference images.

Errors

Your responsibility

The gateway drives your own muse.ai accounts through a browser. Check that this fits Muse's terms before you use it, and accept the risk that an account may be limited or banned. Only generate content you have the right to make, especially when using a real person's face.

Commands

Data location

All runtime state (sessions, license, API keys, usage, media, logs) lives in ~/.muse-gateway — never inside the installed package. Uninstalling keeps data; delete ~/.muse-gateway for a full reset.

• ~/.muse-gateway/server.log — background server log
• ~/.muse-gateway/data/ — sessions, license.json, keys.json, pool, media

Requirements

• Windows 10/11, Node.js ≥ 20.9
• Chrome or Edge installed (used for Muse account login capture)

License

Commercial — UNLICENSED. The gateway API requires a purchased activation key.