fix: point Vercel frontend to HF Space backend

- web/.env.production: VITE_API_BASE_URL=https://vxkyyy-agentIC.hf.space
- web/.gitignore: allow .env.production to be committed
- server/api.py: add LLM pre-flight check in POST /build (503 instead of silent failure)
- requirements.txt: remove unused streamlit/plotly/gdstk deps (faster Docker build)

requirements.txt

@@ -1,13 +1,12 @@
+# ── Core AI agent framework ──────────────────────────────────────────
 crewai
 crewai-tools
 langchain_openai
+
+# ── CLI & Logging ─────────────────────────────────────────────────────
 typer
 rich
 python-dotenv
-streamlit
-pandas
-streamlit-option-menu
-plotly
-streamlit-ace
-gdstk
+
+# ── Server runtime ────────────────────────────────────────────────────
 httpx

server/api.py

@@ -938,6 +938,15 @@ async def trigger_build(req: BuildRequest, profile: dict = Depends(get_current_u
     check_build_allowed(profile)
     byok_key = get_llm_key_for_user(profile)
 
+    # ── LLM pre-flight: fail fast with a clear message ──
+    try:
+        _get_llm(byok_api_key=byok_key)
+    except RuntimeError as e:
+        raise HTTPException(
+            status_code=503,
+            detail=str(e) + " Set NVIDIA_API_KEY in HuggingFace Space secrets.",
+        )
+
     # Sanitize design name — Verilog identifiers cannot start with a digit
     import re as _re
     design_name = req.design_name.strip().lower()

web/.env.production

@@ -0,0 +1,2 @@
+# Production — HuggingFace Space backend
+VITE_API_BASE_URL=https://vxkyyy-agentIC.hf.space

web/.gitignore

@@ -14,6 +14,7 @@ dist-ssr
 .env
 .env.*
 !.env.example
+!.env.production
 
 # Editor directories and files
 .vscode/*