upstream llm_api {
    server llm-api:8100;
    keepalive 32;
}

server {
    listen 80;
    server_name _;

    # Timeouts for slow LLM responses
    proxy_connect_timeout 60s;
    proxy_send_timeout 120s;
    proxy_read_timeout 300s;

    # Health checks (no logging)
    location /health {
        access_log off;
        proxy_pass http://llm_api/health;
        proxy_http_version 1.1;
        proxy_set_header Host $host;
    }

    location /ready {
        access_log off;
        proxy_pass http://llm_api/ready;
        proxy_http_version 1.1;
        proxy_set_header Host $host;
    }

    # API endpoints
    location / {
        proxy_pass http://llm_api;
        proxy_http_version 1.1;

        # Forward client info
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;

        # SSE streaming support
        proxy_set_header Connection '';
        proxy_buffering off;
        proxy_cache off;
        chunked_transfer_encoding off;

        # Large request bodies for long conversations
        client_max_body_size 10M;
    }
}