startup.sh

#!/bin/bash
# ─────────────────────────────────────────────────────────────
# Agent Q-Q (QLAWED-Q) — HF Space Startup v3.1
# MODE: Ollama direct on :7860
#
# VRAM budget (T4-small, 16GB):
#   qwen3:1.7b       1.10GB  triage + chat
#   nomic-embed-text 0.27GB  embeddings / pgvector
#   ──────────────────────────────────────────────
#   Total resident:  ~1.37GB / 16GB
# ─────────────────────────────────────────────────────────────

set -e

echo "╔═════════════════════════════════════════════╗"
echo "║   Agent Q-Q — Ollama Direct Mode           ║"
echo "║   Serving on :7860 (OpenAI-compatible)     ║"
echo "╚═════════════════════════════════════════════╝"
echo ""

# ── Persistent disk (HF Pro) ──────────────────────────────────
if [ -d "/data" ]; then
    echo "✅ Persistent disk at /data"
    mkdir -p /data/ollama-models
    export OLLAMA_MODELS="/data/ollama-models"
else
    echo "⚠️  Ephemeral storage (models re-download on restart)"
fi

# ── Start Ollama on port 7860 ─────────────────────────────────
export OLLAMA_HOST=0.0.0.0:7860
export OLLAMA_KEEP_ALIVE=10m
export OLLAMA_NUM_PARALLEL=2

ollama serve &
OLLAMA_PID=$!

echo "⏳ Waiting for Ollama on :7860..."
for i in {1..40}; do
    if curl -sf http://localhost:7860/api/tags > /dev/null 2>&1; then
        echo "✅ Ollama ready"
        break
    fi
    if [ $i -eq 40 ]; then
        echo "❌ Ollama failed to start"
        exit 1
    fi
    sleep 2
done

# ── Pull models ───────────────────────────────────────────────
echo ""
echo "📦 Pulling models..."

pull_if_missing() {
    local model="$1"
    if ollama show "$model" > /dev/null 2>&1; then
        echo "  ✓ cached: $model"
    else
        echo "  ↓ pulling: $model"
        ollama pull "$model" && echo "  ✅ $model" || echo "  ⚠️  failed: $model (non-fatal)"
    fi
}

pull_if_missing "qwen3:1.7b"
pull_if_missing "nomic-embed-text"

echo ""
echo "📊 Loaded models:"
ollama list
echo ""

echo "═══════════════════════════════════════════════════════════"
echo "✅ Agent Q-Q LIVE"
echo ""
echo "   Public API:  http://0.0.0.0:7860"
echo ""
echo "   Endpoints:"
echo "     GET  /api/tags                  → list models"
echo "     GET  /v1/models                 → OpenAI model list"
echo "     POST /v1/chat/completions       → OpenAI chat"
echo "     POST /api/generate              → Ollama native"
echo "     POST /api/embeddings            → embeddings"
echo ""
echo "   Models resident:"
echo "     qwen3:1.7b        ~1.1GB  chat / triage"
echo "     nomic-embed-text  ~0.3GB  embeddings"
echo "═══════════════════════════════════════════════════════════"

wait $OLLAMA_PID