Update utils/llms.py

utils/llms.py

@@ -6,6 +6,21 @@ import json
 import os
 from g4f.Provider import DeepInfraChat,LambdaChat
 from backup import Client as PerplexityClient,cookies
+import re
+import base64
+import mimetypes
+import urllib.parse
+
+def _ext_from_url(url):
+    try:
+        path = urllib.parse.urlparse(url).path
+        root, ext = os.path.splitext(path)
+        if ext:
+            return ext
+    except Exception:
+        pass
+    return '.jpg'
+
 
 gemini_api_keys=json.loads(os.environ.get("GEMINI_KEY_LIST"))
 groq_api_keys=json.loads(os.environ.get("GROQ_API_KEYS"))
@@ -52,24 +67,97 @@ def chatstream(messages,model,api_keys,tools):
     cunk=""
 
 
+
     if model in ["o3","gpt-4.1",'grok-4','gemini-2.5-pro','claude-sonnet-4-20250514','sonar-pro','r1-1778']:
+        # build the raw message text as before
         raw_perplex_msg = "".join(
             (( (f"[{message['role']}]" ) +  ("(#message)" if message['role']!="system" else "(#instructions)") ) if message['role'] != "assistant"   else "") + f"\n{message['content']}\n\n"
             for message in messages
         )
-        with open("perplexity_messages.txt", "w",encoding='utf-8') as f:
+
+        # --- extract images / image urls into files dict ---
+        files = {}
+        image_counter = 1
+
+        # regexes
+        markdown_image_pattern = re.compile(r'!\[.*?\]\((.*?)\)')
+        image_url_pattern = re.compile(r'https?://\S+\.(?:png|jpg|jpeg|gif|webp|svg)', re.IGNORECASE)
+        data_uri_pattern = re.compile(r'data:(image/[\w.+-]+);base64,([A-Za-z0-9+/=\n\r]+)', re.IGNORECASE)
+
+        for message in messages:
+            # 1) attachments (common OpenAI-like shape)
+            atts = message.get('attachments') or message.get('attachment') or []
+            if isinstance(atts, dict):
+                atts = [atts]
+            if isinstance(atts, list):
+                for att in atts:
+                    if not isinstance(att, dict):
+                        continue
+                    # common keys for urls
+                    url = att.get('url') or att.get('image_url') or att.get('src') or att.get('href')
+                    if url:
+                        ext = _ext_from_url(url)
+                        fname = f'image_{image_counter}{ext}'
+                        files[fname] = url  # pass URL as file value (per your example style)
+                        image_counter += 1
+
+            # 2) content as string -> look for data-URIs, markdown images, plain image URLs
+            content = message.get('content', '') or ''
+            if not isinstance(content, str):
+                try:
+                    content = str(content)
+                except Exception:
+                    content = ''
+
+            # data URIs -> decode and attach raw bytes
+            for m in data_uri_pattern.finditer(content):
+                mime_type, b64 = m.groups()
+                try:
+                    b = base64.b64decode(b64)
+                except Exception:
+                    continue
+                ext = mimetypes.guess_extension(mime_type) or '.bin'
+                fname = f'image_{image_counter}{ext}'
+                files[fname] = b  # raw bytes
+                image_counter += 1
+
+            # markdown images: ![alt](url)
+            for m in markdown_image_pattern.finditer(content):
+                url = m.group(1)
+                ext = _ext_from_url(url)
+                fname = f'image_{image_counter}{ext}'
+                files[fname] = url
+                image_counter += 1
+
+            # plain image URLs in text
+            for m in image_url_pattern.finditer(content):
+                url = m.group(0)
+                ext = _ext_from_url(url)
+                fname = f'image_{image_counter}{ext}'
+                files[fname] = url
+                image_counter += 1
+
+        # also include the textual messages file (so Perplexity gets the full text payload as a file)
+        files['perplexity_messages.txt'] = raw_perplex_msg
+
+        # write the textual file to disk as before (optional, kept for compatibility/debugging)
+        with open("perplexity_messages.txt", "w", encoding='utf-8') as f:
             f.write(raw_perplex_msg)
-        resp = perplexity_cli.search(raw_perplex_msg, mode='reasoning', model=model, sources=[], files={}, stream=True, language='en-US', follow_up=None, incognito=False)
+
+        # run the search, passing files (which may contain URL strings or raw bytes for data URIs)
+        resp = perplexity_cli.search(raw_perplex_msg, mode='reasoning', model=model, sources=[], files=files, stream=True, language='en-US', follow_up=None, incognito=False)
+
+        # stream the response, aggregate into cunk (init if needed)
+        cunk = ""
         for i in resp:
             try:
-                cunk = cunk+(i["blocks"][0]["markdown_block"]["chunks"][0])
+                cunk = cunk + (i["blocks"][0]["markdown_block"]["chunks"][0])
                 if "```json" not in cunk or "```" not in cunk:
                     yield i["blocks"][0]["markdown_block"]["chunks"][0]
-                # print(i["blocks"][0]["markdown_block"]["chunks"][0],end="")
             except Exception as e:
                 print(e)
                 pass
-        yield ("RESULT: "+cunk)
+        yield ("RESULT: " + cunk)
         
     elif model in deepinframodels:
         try:
@@ -345,4 +433,4 @@ def chatstream(messages,model,api_keys,tools):
 # Note:
 # 1)First search and research the web thoroughly end to end.
 # 2)Always Use the website https://collegedunia.com to search about the cut offs.Dont use any other website but only this website.
-# Mandatory:**Give Atleast 10 college names**
+# Mandatory:**Give Atleast 10 college names**