#!/bin/bash
# Sofiia Console — NODA2 local dev startup
# Runs without API key (localhost bypass active), uses Grok by default.
# Usage: ./start-local.sh

set -e
cd "$(dirname "$0")"

# Load root .env if exists (picks up XAI_API_KEY, DEEPSEEK_API_KEY, etc.)
if [ -f "../../.env" ]; then
  set -a
  source "../../.env"
  set +a
fi

# Dev mode — no auth for localhost
export ENV=dev
export PORT=8002

# === Sofiia's HOME is NODA2 (MacBook) ===
# Primary LLM: Grok 4.1 Fast Reasoning (per AGENTS.md)
# XAI_API_KEY, GLM5_API_KEY loaded from root .env above
# Quick tasks: GLM-5
# Local/offline: NODA2 Ollama (qwen3:14b, qwen3.5:35b-a3b, etc.)

# NODA2 local Ollama
export OLLAMA_URL=http://localhost:11434
export SOFIIA_PREFERRED_CHAT_MODEL="${SOFIIA_PREFERRED_CHAT_MODEL:-ollama:qwen3:14b}"
export SOFIIA_OLLAMA_TIMEOUT_SEC="${SOFIIA_OLLAMA_TIMEOUT_SEC:-120}"
export SOFIIA_OLLAMA_VOICE_TIMEOUT_SEC="${SOFIIA_OLLAMA_VOICE_TIMEOUT_SEC:-45}"
export SOFIIA_OLLAMA_KEEP_ALIVE="${SOFIIA_OLLAMA_KEEP_ALIVE:-30m}"
export SOFIIA_OLLAMA_NUM_CTX="${SOFIIA_OLLAMA_NUM_CTX:-8192}"
export SOFIIA_OLLAMA_NUM_THREAD="${SOFIIA_OLLAMA_NUM_THREAD:-8}"
export SOFIIA_OLLAMA_NUM_GPU="${SOFIIA_OLLAMA_NUM_GPU:--1}"
export SOFIIA_OLLAMA_NUM_PREDICT_TEXT="${SOFIIA_OLLAMA_NUM_PREDICT_TEXT:-768}"

# NODA2 memory service
export MEMORY_SERVICE_URL=http://localhost:8000

# NODA1 services (optional — for Router/Telegram context)
export ROUTER_URL=http://144.76.224.179:9102
export GATEWAY_URL=http://144.76.224.179:9300

# Data dir
export SOFIIA_DATA_DIR="$HOME/.sofiia/console-data"
mkdir -p "$SOFIIA_DATA_DIR"

# Activate venv if present
if [ -d "venv" ]; then
  source venv/bin/activate
elif [ -d "../../venv" ]; then
  source ../../venv/bin/activate
fi

echo "🚀 Sofiia Console — http://localhost:8002  (НОДА2, без авторизації)"
echo "   Primary:     Grok 4.1 Fast Reasoning (AGENTS.md)"
echo "   XAI_API_KEY: ${XAI_API_KEY:0:12}..."
echo "   GLM5_API_KEY: ${GLM5_API_KEY:0:12}..."
echo "   OLLAMA_URL:  $OLLAMA_URL (НОДА2 local models)"
echo "   Preferred:   $SOFIIA_PREFERRED_CHAT_MODEL"
echo "   Ollama tune: ctx=$SOFIIA_OLLAMA_NUM_CTX threads=$SOFIIA_OLLAMA_NUM_THREAD gpu=$SOFIIA_OLLAMA_NUM_GPU keep_alive=$SOFIIA_OLLAMA_KEEP_ALIVE"
echo "   Models: qwen3:14b, qwen3.5:35b-a3b, glm-4.7-flash, deepseek-r1:70b..."
echo ""

python -m uvicorn app.main:app --host 127.0.0.1 --port "$PORT" --reload