ForgeHub

A desktop app for orchestrating ComfyUI workflows on Runpod Serverless endpoints — wildcard prompts, QwenVL prompt enhancer, job queue and output gallery. Backend bundled: nothing else to install.

Companion to the workers: runpod-qwen21 (Qwen Image 2.1 T2I/Edit + Pony) · runpod-minimax-h3 (MiniMax H3 video+audio)

📥 Download (desktop app)

Get the latest build from Releases — the backend is bundled, nothing else to install:

Platform	File
macOS (Apple Silicon)	ForgeHub-*-arm64.dmg
Windows	ForgeHub.Setup.*.exe
Linux	ForgeHub-*.AppImage · forgehub-app_*_amd64.deb

Unsigned builds — how to open them:

• macOS: run xattr -dr com.apple.quarantine /Applications/ForgeHub.app (or right-click → Open), or install Sentinel — brew install --cask sentinel — and drop the app on "Allow unsigned app to launch".
• Windows: click "More info → Run anyway".

Then set your Runpod API key in Settings — "Detect from account" lists your endpoints automatically.

⚡ Quick start (from source)

macOS / Linux

git clone https://github.com/huchukato/ForgeHub.git && cd ForgeHub
./start.sh

Windows

git clone https://github.com/huchukato/ForgeHub.git && cd ForgeHub
start.bat

The script handles everything: installs uv if missing, creates .venv and installs the backend (pip install -e .), runs npm install + frontend build, then starts on http://127.0.0.1:8484. On first run it copies .env.example → .env: fill it in (Runpod API key + endpoint id) and relaunch.

Options: ./start.sh --build forces a frontend rebuild · ./start.sh --dev starts backend + Vite dev server (:5173, hot reload).

Prerequisites: only Node.js (for the frontend build) — everything else is handled by uv.

☁️ Execution

Jobs always run on a RunPod serverless endpoint: /run → /status poll → base64 output saved to data/storage and served from /outputs/. Set your API key + endpoint in Settings (or .env) — "Detect from account" lists your endpoints automatically.

Wildcard expansion (__pmp/…__) and runtime selects happen backend-side — the wildcards/ directory is bundled; override with FORGEHUB_WILDCARD_DIRS (colon-separated).

🔧 Environment variables

• FORGEHUB_HOST / FORGEHUB_PORT — backend bind (default 0.0.0.0:8484)
• RUNPOD_API_KEY / RUNPOD_ENDPOINT_ID — serverless endpoint
• FORGEHUB_STORAGE_DIR — output/upload storage (default data/storage)
• FORGEHUB_WILDCARD_DIRS — wildcard dirs (default ./wildcards)
• FORGEHUB_CHAT_LLM_URL / FORGEHUB_CHAT_LLM_MODEL / FORGEHUB_CHAT_LLM_KEY — optional prompt assist (OpenAI-compatible)
• FORGEHUB_CHAT_BASE_URL — optional QwenVL-Mod /qwenvl/chat endpoint
• FORGEHUB_FRONTEND_DIR — frontend dist served from /

See .env.example for the full set.

🖥️ Frontend

cd forgehub-app
npm run dev            # Vite dev server
npm run build          # production build (served by the backend)
npm run electron:dev   # Electron desktop in dev
npm run electron:build # dmg/exe via electron-builder

The packaged desktop app spawns its own bundled backend on 127.0.0.1:8484 — nothing to configure. In electron:dev it loads the Vite dev server, so start the backend too (e.g. ./start.sh).

🧩 Workflows

JSON files in ComfyUI API format (Export API), with an optional .meta.json for metadata/parameters:

{
  "id": "pony-txt2img",
  "name": "Pony XL Txt2Img",
  "category": "image",
  "description": "Generate images with Pony Diffusion XL",
  "tags": ["pony", "sdxl"],
  "parameters": [],
  "outputs": ["image"]
}

Categories: agent, video, image, audio, other.