forge-studio 0.1.0

Web UI for the Forge inference control plane

forge-studio

Web UI for ForgeEngine. FastAPI backend with a React frontend -- chat with Qwen3.6 reasoning models, ingest repos, toggle thinking mode, view token budget.

Install

pip install -e ".[dev]"
cd frontend && npm install && npm run build && cd ..

Run

# Set your inference backend
export FORGE_MODEL=Qwen3.6-27B
export FORGE_BASE_URL=http://localhost:8000/v1

# Start the server (binds to 127.0.0.1 by default)
forge-studio --port 8400

Open http://localhost:8400 in your browser.

Development mode

Run backend and frontend separately for hot reloading:

# Terminal 1: backend with auto-reload
forge-studio --port 8400 --dev

# Terminal 2: frontend dev server (proxies API to backend)
cd frontend && npm run dev

The Vite dev server runs on port 5173 and proxies /api requests to the backend on port 8400.

Configuration

Environment variables:

Variable	Default	Description
FORGE_MODEL	Qwen3.6-27B	Model name
FORGE_BASE_URL	http://localhost:8000/v1	Inference API endpoint
FORGE_API_KEY	EMPTY	API key for the endpoint
FORGE_BACKEND	(auto)	Backend type: vllm, sglang, dashscope
FORGE_GPU_ID	(none)	GPU ID for MTP recommendation

API

Method	Endpoint	Description
POST	/api/chat	Non-streaming chat
WS	/api/chat/stream	Streaming chat via WebSocket
GET	/api/status	Engine status and budget info
GET	/api/messages	Conversation history
POST	/api/ingest	Ingest a git repository
POST	/api/ingest-local	Ingest a local repo path
GET	/api/mtp	MTP recommendation
POST	/api/thinking-mode	Set thinking mode (think/no_think)
POST	/api/backend	Switch backend type

Tests

pytest -x -q

License

Apache 2.0