fix: remove Ollama+model from image, shrink from 4GB to 500MB for HF stability

Dockerfile

@@ -1,16 +1,16 @@
-# Dockerfile for Hugging Face Spaces
+# ☁️ Dockerfile for Hugging Face Spaces — Cloud-Only (Lightweight)
+# Image size: ~500MB (no Ollama, no local models)
+# All LLM inference is handled by Groq API and HF Inference API via USE_HF_CLOUD=1
 FROM python:3.11-slim
 
 # Prevent Python from writing pyc files and buffering stdout/stderr
 ENV PYTHONDONTWRITEBYTECODE=1
 ENV PYTHONUNBUFFERED=1
-ENV USE_HF_CLOUD=1
 
-# Pre-set Ollama model directory so the build-time pull goes to the right place
-ENV OLLAMA_MODELS=/app/ollama_models
-ENV HOME=/app
+# Cloud mode: all LLM calls go to Groq / HF Inference — no local Ollama needed
+ENV USE_HF_CLOUD=1
 
-# Install system dependencies
+# Install system dependencies (Tesseract for OCR fallback only)
 RUN apt-get update && apt-get install -y \
     curl \
     zstd \
@@ -20,31 +20,6 @@ RUN apt-get update && apt-get install -y \
     libpq-dev \
     && rm -rf /var/lib/apt/lists/*
 
-# Install latest Ollama via official install script
-# (replaces the pinned v0.5.11 which cannot pull qwen3.5:2b)
-RUN curl -fsSL https://ollama.ai/install.sh | sh
-
-# Pre-bake qwen3.5:2b vision model into the image at build time.
-# The model will be stored in /app/ollama_models and available
-# immediately when the container starts — no runtime pull needed.
-RUN mkdir -p /app/ollama_models && \
-    OLLAMA_MODELS=/app/ollama_models ollama serve > /tmp/ollama_build.log 2>&1 & \
-    OLLAMA_PID=$! && \
-    echo "[BUILD] Waiting for Ollama to start (PID=$OLLAMA_PID)..." && \
-    for i in $(seq 1 30); do \
-        if curl -s http://localhost:11434/api/tags > /dev/null 2>&1; then \
-            echo "[BUILD] Ollama is ready."; break; \
-        fi; \
-        echo "[BUILD] Waiting... ($i/30)"; \
-        sleep 2; \
-    done && \
-    echo "[BUILD] Pulling qwen3.5:2b into image layers..." && \
-    OLLAMA_MODELS=/app/ollama_models ollama pull qwen3.5:2b && \
-    echo "[BUILD] Model pull complete. Stopping build-time Ollama..." && \
-    kill $OLLAMA_PID 2>/dev/null || true && \
-    wait $OLLAMA_PID 2>/dev/null || true && \
-    echo "[BUILD] qwen3.5:2b pre-baked successfully."
-
 WORKDIR /app
 
 # Install Python dependencies

start_hf.sh

@@ -1,45 +1,10 @@
 #!/bin/bash
 
-# Cloud-Optimized Startup Script for Hugging Face Spaces
-# Model is pre-baked into the image — no runtime pull needed.
+# ☁️ Cloud-Only Startup Script for Hugging Face Spaces
+# All LLM calls are routed to Groq and HF Inference APIs via USE_HF_CLOUD=1
+# Ollama is NOT used — image is kept lightweight to prevent HF storage purges.
 
-# 0. Start Ollama server
-echo "[HF-INIT] Starting Ollama server..."
-export HOME=/app
-export OLLAMA_HOST=0.0.0.0:11434
-export OLLAMA_MODELS=/app/ollama_models
-
-ollama serve > /app/ollama.log 2>&1 &
-
-# Wait for Ollama to be ready
-echo "[HF-INIT] Waiting for Ollama to start..."
-for i in {1..20}; do
-    if curl -s http://localhost:11434/api/tags > /dev/null 2>&1; then
-        echo "[HF-INIT] Ollama is ready."
-        break
-    fi
-    if [ $i -eq 10 ]; then
-        echo "[HF-INIT] Still waiting — Ollama log:"
-        cat /app/ollama.log
-    fi
-    echo "[HF-INIT] Waiting... ($i/20)"
-    sleep 2
-done
-
-# Verify the pre-baked model is present before starting the app.
-# If it is missing for any reason, pull it now (blocking, so app waits).
-echo "[HF-INIT] Verifying qwen3.5:2b is available..."
-if ollama list 2>/dev/null | grep -q "qwen3.5:2b"; then
-    echo "[HF-INIT] qwen3.5:2b confirmed present."
-else
-    echo "[HF-INIT] WARNING: qwen3.5:2b not found in pre-baked layers. Pulling now..."
-    ollama pull qwen3.5:2b
-    if [ $? -eq 0 ]; then
-        echo "[HF-INIT] Pull succeeded."
-    else
-        echo "[HF-INIT] ERROR: Pull failed. Vision Q&A will not work."
-    fi
-fi
+echo "[HF-INIT] ===== Application Startup at $(date -u '+%Y-%m-%d %H:%M:%S') ====="
 
 # 1. Start Backend Worker
 echo "[HF-INIT] Starting Backend Worker on port 8700..."
@@ -53,5 +18,7 @@ python -m uvicorn "Agent.mcp_server:app" --host 0.0.0.0 --port 8701 &
 echo "[HF-INIT] Starting Web UI on port 7860..."
 python -m uvicorn "web.app:app" --host 0.0.0.0 --port 7860 &
 
+echo "[HF-INIT] All services started. Container is live."
+
 # Keep the container alive
 wait