Add abuse mitigation: 10 msg/min, 50 msg/session, 2000 char limit, 5000 daily global cap

app.py

@@ -15,6 +15,10 @@ from __future__ import annotations
 
 import io
 import os
+import time
+import threading
+from collections import deque
+from datetime import datetime, timezone
 
 import gradio as gr
 import requests
@@ -31,6 +35,33 @@ DEMO_BANNER = (
     "128B parameters · 278K context · KVKK-resident"
 )
 
+# ── Abuse mitigation knobs ─────────────────────────────────────────
+RATE_PER_MIN_PER_SESSION = int(os.environ.get("RATE_PER_MIN", "10"))
+MAX_PER_SESSION = int(os.environ.get("MAX_PER_SESSION", "50"))
+MAX_PROMPT_CHARS = int(os.environ.get("MAX_PROMPT_CHARS", "2000"))
+DAILY_CAP_GLOBAL = int(os.environ.get("DAILY_CAP", "5000"))
+
+# ── Global daily counter (thread-safe) ─────────────────────────────
+_global_lock = threading.Lock()
+_global_counter = {"date": "", "count": 0}
+
+
+def _utc_today() -> str:
+    return datetime.now(timezone.utc).strftime("%Y-%m-%d")
+
+
+def _check_and_inc_daily() -> tuple[bool, int]:
+    """Increment the global daily counter; return (allowed, current_count)."""
+    today = _utc_today()
+    with _global_lock:
+        if _global_counter["date"] != today:
+            _global_counter["date"] = today
+            _global_counter["count"] = 0
+        if _global_counter["count"] >= DAILY_CAP_GLOBAL:
+            return False, _global_counter["count"]
+        _global_counter["count"] += 1
+        return True, _global_counter["count"]
+
 PLACEHOLDER_MSG = (
     "🔒 The interactive chat is being activated.\n\n"
     "While the demo is finalised, you can already:\n"
@@ -105,31 +136,75 @@ def send_with_image(chat_id: str, message: str, image: Image.Image) -> str:
     return r.json().get("message", "")
 
 
-# ── Gradio handler with session-scoped chat_id ─────────────────────
-def chat(prompt, image, history, chat_id_state):
+# ── Gradio handler with session-scoped chat_id + rate limiting ────
+def chat(prompt, image, history, chat_id_state, ts_log_state, count_state):
+    """
+    ts_log_state: deque of recent timestamps (sliding 60s window) for per-minute rate
+    count_state: total messages in this session
+    """
     history = history or []
     chat_id = chat_id_state or ""
+    ts_log = ts_log_state or deque()
+    count = count_state or 0
 
     if not prompt.strip():
-        return history, "", chat_id
+        return history, "", chat_id, ts_log, count
+
+    # 1) Prompt length limit
+    if len(prompt) > MAX_PROMPT_CHARS:
+        history.append((prompt[:200] + " […]", (
+            f"⚠️ Prompt too long ({len(prompt)} chars). "
+            f"Max {MAX_PROMPT_CHARS} chars per message in this demo. "
+            f"For longer contexts, use the production API at aigency.dev."
+        )))
+        return history, "", chat_id, ts_log, count
+
+    # 2) Per-session total
+    if count >= MAX_PER_SESSION:
+        history.append((prompt, (
+            f"⚠️ Session limit reached ({MAX_PER_SESSION} messages). "
+            f"Refresh the page to start a new session, "
+            f"or contact info@e-cloud.web.tr · ai@aigency.dev for production access."
+        )))
+        return history, "", chat_id, ts_log, count
+
+    # 3) Per-minute rate (sliding window)
+    now = time.time()
+    while ts_log and now - ts_log[0] > 60:
+        ts_log.popleft()
+    if len(ts_log) >= RATE_PER_MIN_PER_SESSION:
+        wait = int(60 - (now - ts_log[0]))
+        history.append((prompt, (
+            f"⚠️ Slow down — max {RATE_PER_MIN_PER_SESSION} messages/minute. "
+            f"Try again in {wait}s."
+        )))
+        return history, "", chat_id, ts_log, count
+
+    # 4) Global daily cap
+    allowed, daily_count = _check_and_inc_daily()
+    if not allowed:
+        history.append((prompt, (
+            f"⚠️ The demo has reached today's global limit ({DAILY_CAP_GLOBAL} requests). "
+            f"It resets at 00:00 UTC. For uninterrupted access, contact "
+            f"info@e-cloud.web.tr · ai@aigency.dev."
+        )))
+        return history, "", chat_id, ts_log, count
 
     if not API_TOKEN:
         history.append((prompt, PLACEHOLDER_MSG))
-        return history, "", chat_id
+        return history, "", chat_id, ts_log, count
 
+    # 5) Actual API call
     try:
         if not chat_id:
-            # Open a new conversation; the seed message can be the user's first prompt
             if image is None:
                 cid, answer = new_chat(prompt)
                 chat_id = cid
             else:
-                # newChat doesn't accept attachments — open with a brief seed, then send image
                 cid, _ = new_chat("Bir görsel inceleyeceksin.")
                 chat_id = cid
                 answer = send_with_image(chat_id, prompt, image)
         else:
-            # Continue existing conversation
             if image is None:
                 answer = send_message(chat_id, prompt)
             else:
@@ -137,12 +212,13 @@ def chat(prompt, image, history, chat_id_state):
     except Exception as e:
         answer = f"Error: {e}"
 
+    ts_log.append(now)
     history.append((prompt, answer))
-    return history, "", chat_id
+    return history, "", chat_id, ts_log, count + 1
 
 
 def reset_session():
-    return [], "", ""
+    return [], "", "", deque(), 0
 
 
 def make_leaderboard():
@@ -175,6 +251,8 @@ Responses are served live by the production API.
 with gr.Blocks(title="AIGENCY V4 Demo", theme=gr.themes.Soft()) as demo:
     gr.Markdown(f"# AIGENCY V4\n\n*{DEMO_BANNER}*")
     chat_id_state = gr.State("")
+    ts_log_state = gr.State(lambda: deque())
+    count_state = gr.State(0)
 
     with gr.Tab("Chat"):
         with gr.Row():
@@ -195,9 +273,18 @@ with gr.Blocks(title="AIGENCY V4 Demo", theme=gr.themes.Soft()) as demo:
         with gr.Row():
             send = gr.Button("Send", variant="primary")
             clear = gr.Button("New conversation")
-        send.click(chat, [msg, img, chatbot, chat_id_state], [chatbot, msg, chat_id_state])
-        msg.submit(chat, [msg, img, chatbot, chat_id_state], [chatbot, msg, chat_id_state])
-        clear.click(reset_session, [], [chatbot, msg, chat_id_state])
+        gr.Markdown(
+            f"*Demo limits: ≤ {MAX_PROMPT_CHARS} chars/message · "
+            f"{RATE_PER_MIN_PER_SESSION} msg/min · "
+            f"{MAX_PER_SESSION} msg/session · "
+            f"{DAILY_CAP_GLOBAL} requests/day globally. "
+            f"For unlimited production access: info@e-cloud.web.tr · ai@aigency.dev*"
+        )
+        chat_inputs = [msg, img, chatbot, chat_id_state, ts_log_state, count_state]
+        chat_outputs = [chatbot, msg, chat_id_state, ts_log_state, count_state]
+        send.click(chat, chat_inputs, chat_outputs)
+        msg.submit(chat, chat_inputs, chat_outputs)
+        clear.click(reset_session, [], [chatbot, msg, chat_id_state, ts_log_state, count_state])
 
     with gr.Tab("Benchmark Leaderboard"):
         gr.Markdown(