eve / app.py
Chandima Prabhath
Revert "Fix send_status_image to use correct key for uploaded file URL in response"
87a21e8
raw
history blame
21 kB
import os
import threading
import requests
import logging
import queue
import re
import json
import time
import random
from fastapi import FastAPI, Request, HTTPException
from fastapi.responses import PlainTextResponse, JSONResponse
from FLUX import generate_image
from VoiceReply import generate_voice_reply
from polLLM import generate_llm
# Configure logging
logging.basicConfig(level=logging.DEBUG, format="%(asctime)s [%(levelname)s] %(message)s")
# Env vars
GREEN_API_URL = os.getenv("GREEN_API_URL")
GREEN_API_MEDIA_URL = os.getenv("GREEN_API_MEDIA_URL")
GREEN_API_TOKEN = os.getenv("GREEN_API_TOKEN")
GREEN_API_ID_INSTANCE = os.getenv("GREEN_API_ID_INSTANCE")
WEBHOOK_AUTH_TOKEN = os.getenv("WEBHOOK_AUTH_TOKEN")
BOT_STATUS_CHAT = "120363312903494448@g.us" # Chat ID for system messages
image_dir = "/tmp/images"
audio_dir = "/tmp/audio"
# Ensure necessary directories exist
os.makedirs(image_dir, exist_ok=True)
os.makedirs(audio_dir, exist_ok=True)
if not all([GREEN_API_URL, GREEN_API_TOKEN, GREEN_API_ID_INSTANCE, WEBHOOK_AUTH_TOKEN]):
raise ValueError("Environment variables are not set properly")
# Queues & in‑memory stores
task_queue = queue.Queue()
trivia_store = {} # chat_id β†’ {"question":…, "answer":…}
polls = {} # chat_id β†’ {"question":…, "options": […], "votes": {1: 0, ...}, "voters": {jid: opt}}
app = FastAPI()
# Global inactivity tracker
last_message_time = time.time()
# --- Inactivity Monitor ---
def inactivity_monitor():
global last_message_time
while True:
time.sleep(60) # check every minute
if time.time() - last_message_time >= 300: # 5 minutes inactivity
if BOT_STATUS_CHAT:
reminder = "⏰ I haven't heard from you in a while! I'm still here if you need anything."
send_message("inactivity", BOT_STATUS_CHAT, reminder)
last_message_time = time.time()
threading.Thread(target=inactivity_monitor, daemon=True).start()
# --- Background Worker ---
def worker():
while True:
task = task_queue.get()
try:
typ = task["type"]
mid = task["message_id"]
cid = task["chat_id"]
if typ == "image":
handle_image_generation(mid, cid, task["prompt"])
elif typ == "audio":
response_audio(mid, cid, task["prompt"])
except Exception as e:
logging.error(f"Error processing {task}: {e}")
finally:
task_queue.task_done()
threading.Thread(target=worker, daemon=True).start()
# --- send helpers ---
def send_message_to_chat(to_number, message, retries=3):
chat_id = to_number if to_number.endswith("@g.us") else to_number
url = f"{GREEN_API_URL}/waInstance{GREEN_API_ID_INSTANCE}/sendMessage/{GREEN_API_TOKEN}"
payload = {"chatId": chat_id, "message": message}
for i in range(retries):
try:
r = requests.post(url, json=payload)
r.raise_for_status()
return r.json()
except requests.RequestException as e:
if i == retries - 1:
logging.error("send_message_to_chat failed: %s", str(e))
return {"error": str(e)}
def send_message(message_id, to_number, message, retries=3):
chat_id = to_number if to_number.endswith("@g.us") else to_number
url = f"{GREEN_API_URL}/waInstance{GREEN_API_ID_INSTANCE}/sendMessage/{GREEN_API_TOKEN}"
payload = {"chatId": chat_id, "message": message, "quotedMessageId": message_id}
for i in range(retries):
try:
r = requests.post(url, json=payload)
r.raise_for_status()
return r.json()
except requests.RequestException as e:
if i == retries - 1:
return {"error": str(e)}
def send_image(message_id, to_number, image_path, caption="Here you go!", retries=3):
chat_id = to_number if to_number.endswith("@g.us") else to_number
url = f"{GREEN_API_MEDIA_URL}/waInstance{GREEN_API_ID_INSTANCE}/sendFileByUpload/{GREEN_API_TOKEN}"
payload = {"chatId": chat_id, "caption": caption, "quotedMessageId": message_id}
files = [("file", ("image.jpg", open(image_path, "rb"), "image/jpeg"))]
for i in range(retries):
try:
r = requests.post(url, data=payload, files=files)
r.raise_for_status()
return r.json()
except requests.RequestException as e:
if i == retries - 1:
return {"error": str(e)}
def send_audio(message_id, to_number, audio_path, retries=3):
logging.debug("send_audio")
chat_id = to_number if to_number.endswith("@g.us") else to_number
if not os.path.exists(audio_path):
logging.debug(f"Missing audio: {audio_path}")
url = f"{GREEN_API_MEDIA_URL}/waInstance{GREEN_API_ID_INSTANCE}/sendFileByUpload/{GREEN_API_TOKEN}"
payload = {"chatId": chat_id, "caption": "Here is your voice reply!", "quotedMessageId": message_id}
try:
with open(audio_path, "rb") as f:
files = [("file", ("audio.mp3", f, "audio/mpeg"))]
for i in range(retries):
try:
r = requests.post(url, data=payload, files=files)
r.raise_for_status()
return r.json()
except requests.RequestException as e:
if i == retries - 1:
return {"error": str(e)}
except Exception as e:
return {"error": str(e)}
# --- New helper: send WhatsApp status image ---
def send_status_image(image_path, caption="Status Update", retries=3):
"""
Uploads an image file using Green API's uploadFile endpoint to obtain a URL,
then sends the WhatsApp status update with that URL using sendMediaStatus.
"""
# Step 1: Upload the image to get a file URL
upload_url = f"{GREEN_API_MEDIA_URL}/waInstance{GREEN_API_ID_INSTANCE}/uploadFile/{GREEN_API_TOKEN}"
logging.debug("Uploading image to Green API: %s", upload_url)
try:
with open(image_path, "rb") as f:
files = {"file": f}
upload_response = requests.post(upload_url, files=files)
upload_response.raise_for_status()
upload_data = upload_response.json()
# Adjust key as per API response. Here we try "url" or "fileUrl".
file_url = upload_data.get("url") or upload_data.get("fileUrl")
if not file_url:
logging.error("Upload failed: no file URL returned.")
return {"error": "No file URL returned from upload"}
logging.info("Image uploaded successfully. URL: %s", file_url)
except Exception as e:
logging.error("Error uploading image: %s", e)
return {"error": str(e)}
# Step 2: Use the file URL to send the status update
status_url = f"{GREEN_API_MEDIA_URL}/waInstance{GREEN_API_ID_INSTANCE}/sendMediaStatus/{GREEN_API_TOKEN}"
payload = {
"urlFile": file_url,
"caption": caption,
"fileName": os.path.basename(image_path)
}
for i in range(retries):
try:
r = requests.post(status_url, json=payload)
r.raise_for_status()
logging.info("Status image sent successfully.")
return r.json()
except requests.RequestException as e:
if i == retries - 1:
logging.error("send_status_image failed: %s", str(e))
return {"error": str(e)}
# --- core response functions ---
def response_text(message_id, chat_id, prompt):
try:
msg = generate_llm(prompt)
send_message(message_id, chat_id, msg)
except Exception:
send_message(message_id, chat_id, "Error processing your request.")
def response_audio(message_id, chat_id, prompt):
logging.debug("response_audio prompt=%s", prompt)
try:
result = generate_voice_reply(prompt, model="openai-audio", voice="coral", audio_dir=audio_dir)
if result and result[0]:
audio_path, _ = result
send_audio(message_id, chat_id, audio_path)
if os.path.exists(audio_path):
os.remove(audio_path)
else:
response_text(message_id, chat_id, prompt)
except Exception as e:
logging.debug("audio error: %s", e)
send_message(message_id, chat_id, "Error generating audio. Try again later.")
def handle_image_generation(message_id, chat_id, prompt):
for i in range(4):
try:
img, path, ret_prompt, url = generate_image(prompt, message_id, message_id, image_dir)
if img:
formatted_ret_prompt = "\n\n".join(
f"_{paragraph.strip()}_" for paragraph in ret_prompt.split("\n\n") if paragraph.strip()
)
send_image(
message_id,
chat_id,
path,
caption=f"✨ Image ready: {url}\n>{chr(8203)} {formatted_ret_prompt}"
)
else:
send_message(message_id, chat_id, "Image generation failed.")
break # exit on success
except Exception as e:
logging.error("Error in handle_image_generation: %s", e)
send_message(message_id, chat_id, "Error generating image.")
# --- New background thread: WhatsApp Status Updater ---
def whatsapp_status_updater():
"""
Every 2 minutes, this function uses the LLM to generate a random creative image prompt,
generates an image from it, uploads the image to get a URL, and sends it as a WhatsApp status update.
"""
while True:
try:
# Generate a random image prompt via LLM
random_prompt = generate_llm("Generate a creative and random image prompt for a WhatsApp status update.")
logging.info("Random status prompt: %s", random_prompt)
# Generate image from the prompt; using "status" as dummy identifiers.
img, path, ret_prompt, url = generate_image(random_prompt, "status", "status", image_dir)
if img and os.path.exists(path):
caption = f"Status: {random_prompt}"
send_status_image(path, caption)
os.remove(path)
else:
logging.error("Image generation for status failed.")
except Exception as e:
logging.error("Error in whatsapp_status_updater: %s", e)
# Sleep for 2 minutes
time.sleep(120)
# Start the WhatsApp status updater thread
threading.Thread(target=whatsapp_status_updater, daemon=True).start()
# --- Startup Message ---
def send_startup_message():
if BOT_STATUS_CHAT:
startup_msg = "🌟 Hi! I'm Eve, your friendly AI assistant. I'm now live and ready to help with images, voice replies, and more!"
resp = send_message_to_chat(BOT_STATUS_CHAT, startup_msg)
if "error" in resp:
logging.error("Startup message failed: %s", resp["error"])
else:
logging.warning("BOT_STATUS_CHAT is not set; startup message not sent.")
help_text = (
"πŸ€– *Hi there, I'm Eve!* Here are the commands you can use:\n\n"
"β€’ */help* – _Show this help message._\n"
"β€’ */summarize <text>* – _Get a quick summary of your text._\n"
"β€’ */translate <language>|<text>* – _Translate text to your chosen language._\n"
"β€’ */joke* – _Enjoy a random, funny joke._\n"
"β€’ */weather <location>* – _Get the current weather for a location._\n"
"β€’ */inspire* – _Receive a short inspirational quote._\n"
"β€’ */trivia* – _Start a new trivia question._\n"
"β€’ */answer [your answer]* – _Reveal the trivia answer or check your answer if provided._\n"
"β€’ */meme <text>* – _Generate a fun meme image._\n"
"β€’ */poll <Question>|<Option1>|<Option2>|…* – _Create a poll._\n"
"β€’ */results* – _See current poll results._\n"
"β€’ */endpoll* – _End the poll and show final results._\n"
"β€’ */gen <prompt>* – _Generate an image from your prompt._\n\n"
"Send any other text and I'll reply with a voice message. I'm here to help, so don't hesitate to ask!"
)
# --- Webhook ---
@app.post("/whatsapp")
async def whatsapp_webhook(request: Request):
global last_message_time
last_message_time = time.time()
auth = request.headers.get("Authorization", "").strip()
if auth != f"Bearer {WEBHOOK_AUTH_TOKEN}":
raise HTTPException(403, "Unauthorized")
try:
data = await request.json()
chat_id = data["senderData"]["chatId"]
print(f"New message from chat ID: {chat_id}")
except:
return JSONResponse({"error": "Invalid JSON"}, status_code=400)
if data.get("typeWebhook") != "incomingMessageReceived":
return {"success": True}
logging.debug("recv: %s", data)
sd = data["senderData"]
chat = sd["chatId"]
mid = data["idMessage"]
sender_jid = sd.get("sender")
md = data.get("messageData", {})
if md.get("typeMessage") == "quotedMessage" or "quotedMessage" in md:
logging.debug("skip native quotedMessage")
return {"success": True}
if "textMessageData" in md:
body = md["textMessageData"].get("textMessage", "").strip()
ctx = md["textMessageData"].get("contextInfo", {})
elif "extendedTextMessageData" in md:
body = md["extendedTextMessageData"].get("text", "").strip()
ctx = md["extendedTextMessageData"].get("contextInfo", {})
else:
return {"success": True}
if ctx.get("mentionedJid") or ctx.get("mentionedJidList"):
return {"success": True}
if chat.endswith("@g.us") and re.search(r"@\d+", body):
return {"success": True}
low = body.lower()
# --- New Commands ---
if low == "/help":
send_message(mid, chat, help_text)
return {"success": True}
if low.startswith("/summarize "):
txt = body[len("/summarize "):].strip()
summary = generate_llm(f"Summarize this text in one short paragraph:\n\n{txt}")
send_message(mid, chat, summary)
return {"success": True}
if low.startswith("/translate "):
part = body[len("/translate "):]
if "|" not in part:
send_message(mid, chat, "Please use `/translate <language>|<text>`")
else:
lang, txt = part.split("|", 1)
resp = generate_llm(f"Translate the following into {lang.strip()}:\n\n{txt.strip()}")
send_message(mid, chat, resp)
return {"success": True}
if low == "/joke":
try:
joke = requests.get("https://official-joke-api.appspot.com/random_joke", timeout=5).json()
send_message(mid, chat, f"{joke['setup']}\n\n{joke['punchline']}")
except:
send_message(mid, chat, generate_llm("Tell me a short, funny joke."))
return {"success": True}
if low.startswith("/weather "):
loc = body[len("/weather "):].strip().replace(" ", "+")
try:
w = requests.get(f"http://sl.wttr.in/{loc}?format=4", timeout=5).text
send_message(mid, chat, w)
except:
send_message(mid, chat, "Could not fetch weather.")
return {"success": True}
if low == "/inspire":
quote = generate_llm("Give me a short inspirational quote.")
send_message(mid, chat, f"✨ {quote}")
return {"success": True}
# TRIVIA
if low == "/trivia":
randomSeed = random.randint(0, 9999999)
raw = generate_llm(
(f"Generate a unique and random trivia question and answer based on this random seed {randomSeed} to make it unique in JSON format. "
"The output should strictly follow this example format without extra text:\n\n"
"{\"question\": \"What is the capital of France?\", \"answer\": \"Paris\"}")
)
def extract_json(text):
import re
match = re.search(r"```(?:json)?\s*(\{.*?\})\s*```", text, re.DOTALL)
if match:
return match.group(1)
return text
try:
json_text = extract_json(raw)
obj = json.loads(json_text)
if "question" in obj and "answer" in obj:
trivia_store[chat] = obj
send_message(mid, chat, f"❓ {obj['question']}\nReply with `/answer` followed by your answer (if you want to check it) or just `/answer` to reveal the correct answer.")
else:
raise ValueError("Missing expected keys.")
except Exception as e:
logging.error("Trivia JSON parse error: %s, raw response: %s", e, raw)
send_message(mid, chat, "Failed to generate trivia. Please try again.")
return {"success": True}
if low.startswith("/answer"):
user_response = body[len("/answer"):].strip()
if chat in trivia_store:
correct_answer = trivia_store[chat]["answer"]
question = trivia_store[chat]["question"]
if user_response:
eval_prompt = (
f"Question: {question}\n"
f"Correct Answer: {correct_answer}\n"
f"User Answer: {user_response}\n"
"Is the user's answer correct? Respond with 'Correct' if yes, or 'Incorrect' if not, and explain briefly."
)
verdict = generate_llm(eval_prompt)
send_message(mid, chat, f"πŸ’‘ {verdict}")
else:
send_message(mid, chat, f"πŸ’‘ Answer: {correct_answer}")
trivia_store.pop(chat, None)
else:
send_message(mid, chat, "No active trivia. Send `/trivia` to start one.")
return {"success": True}
if low.startswith("/meme "):
txt = body[len("/meme "):].strip()
send_message(mid, chat, "🎨 Generating your meme...")
task_queue.put({
"type": "image",
"message_id": mid,
"chat_id": chat,
"prompt": f"meme template with text: {txt}"
})
return {"success": True}
if low.startswith("/poll "):
parts = body[len("/poll "):].split("|")
if len(parts) < 3:
send_message(mid, chat, "Please use `/poll Question|Option1|Option2|...`")
else:
q = parts[0].strip()
opts = [p.strip() for p in parts[1:]]
votes = {i+1: 0 for i in range(len(opts))}
polls[chat] = {"question": q, "options": opts, "votes": votes, "voters": {}}
txt = f"πŸ“Š *Poll:* {q}\n" + "\n".join(
f"{i+1}. {opt}" for i, opt in enumerate(opts)
) + "\n\nReply with the *option number* to vote."
send_message(mid, chat, txt)
return {"success": True}
if chat in polls and body.isdigit():
n = int(body)
p = polls[chat]
if 1 <= n <= len(p["options"]):
prev = p["voters"].get(sender_jid)
if prev:
p["votes"][prev] -= 1
p["votes"][n] += 1
p["voters"][sender_jid] = n
send_message(mid, chat, f"βœ… Vote recorded: {p['options'][n-1]}")
return {"success": True}
if low == "/results":
if chat in polls:
p = polls[chat]
txt = f"πŸ“Š *Results:* {p['question']}\n" + "\n".join(
f"{i}. {opt}: {p['votes'][i]}" for i, opt in enumerate([""] + p["options"]) if i > 0
)
send_message(mid, chat, txt)
else:
send_message(mid, chat, "No active poll.")
return {"success": True}
if low == "/endpoll":
if chat in polls:
p = polls.pop(chat)
txt = f"πŸ“Š *Final Results:* {p['question']}\n" + "\n".join(
f"{i}. {opt}: {p['votes'][i]}" for i, opt in enumerate([""] + p["options"]) if i > 0
)
send_message(mid, chat, txt)
else:
send_message(mid, chat, "No active poll.")
return {"success": True}
if low.startswith("/gen"):
prompt = body[len("/gen"):].strip()
if not prompt:
send_message(mid, chat, "Please use `/gen <prompt>` to generate an image.")
else:
send_message(mid, chat, "✨ Your image is being generated. Please wait...")
task_queue.put({
"type": "image",
"message_id": mid,
"chat_id": chat,
"prompt": prompt
})
return {"success": True}
task_queue.put({
"type": "audio",
"message_id": mid,
"chat_id": chat,
"prompt": body
})
return {"success": True}
@app.get("/", response_class=PlainTextResponse)
def index():
return "Server is running!"
if __name__ == "__main__":
send_startup_message()
import uvicorn
uvicorn.run(app, host="0.0.0.0", port=7860)