Update app.py

app.py

@@ -26,6 +26,9 @@ import numpy as np
 from pylatexenc.latex2text import LatexNodes2Text
 import requests
 from urllib.parse import quote
+import geo_distance
+import geo_locate
+
 
 load_dotenv(override=True)
 key = os.getenv('OPENAI_API_KEY')
@@ -37,6 +40,7 @@ DEEPSEEK_KEY=os.getenv('DEEPSEEK_KEY')
 GROQ_KEY=os.getenv('GROQ_KEY')
 BRAVE_KEY=os.getenv('BRAVE_KEY')
 BRAVE_SEARCH_KEY=os.getenv('BRAVE_SEARCH_KEY')
+LOCATIONID_KEY=os.getenv('LOCATIONID_KEY')
 
 site = os.getenv('SITE')
 if site == 'local':
@@ -64,6 +68,12 @@ special_chat_types = ['math', 'logic']
 
 news_interval_choices = [("Day", "pd"), ("Week", "pw"), ("Month", "pm"), ("Year", "py")]
 
+def get_distance(addr1, addr2):
+    (lat1, lon1) = geo_locate.get_geo_coords(addr1, LOCATIONID_KEY)
+    (lat2, lon2) = geo_locate.get_geo_coords(addr2, LOCATIONID_KEY)
+    distance = geo_distance.great_circle_distance_miles(lat1, lon1, lat2, lon2)
+    return distance
+
 def get_openai_file(file_id, container_id):
     url = f'https://api.openai.com/v1/containers/{container_id}/files/{file_id}/content'
     headers= {"Authorization": "Bearer " + key}
@@ -753,14 +763,8 @@ def updatePassword(txt):
 #             ref = len(txt[ref:loc]) + len(frag)
 #     return txt
 
-def get_response(prompt, previous_response_id, container_id):
-    if prompt.casefold().find('code') > -1:
-        instructions = '''
-You are a personal math tutor. When asked a math question, 
-write and run code using the python tool to answer the question.
-'''
-    else:
-        instructions = '''
+def get_response(inputs, previous_response_id, container_id, image_file):
+    instructions = '''
 You are a helpful assistant who knows how to browse the web for info and to write and run python
 code. 
 '''
@@ -769,15 +773,47 @@ Do not use latex for math expressions in text output.
 If a chart, table or plot is produced, return it as an image.
 If a powerpoint slide is created, return it as an image but do not offer a download link.
 If the user asks you to output a file, You must include the file you generate in the annotation
-of the output text using a markdown url link in this format as an example: sandbox:/mnt/data/int100.txt
+of the output text.
 '''
+    if image_file != '':
+        with open(image_file, 'rt') as fp:
+            b64data = fp.read()
+        inputs.append(
+            {
+                "role": "user",
+                 "content": [
+                     {
+                         "type": "input_image",
+                         "image_url": f'data:image/jpeg;base64, {b64data}',
+                    }
+                ]
+            }
+        )
+
     response = Client().responses.create(
         model= "gpt-5-mini",      #"gpt-5-mini", "o4-mini",
         tools=[{ "type": "web_search_preview" },
-               { "type": "code_interpreter", "container": container_id}],  # {'type': 'auto'}}],
+                { "type": "code_interpreter", "container": container_id},    #{'type': 'auto'}},
+            {"type": "function", "name": "get_distance",
+              "description": "get calculated straight-line (great-circle) distance between two locations or addresses.",
+              "parameters": {
+                  "type": "object", "properties": {
+                      "addr1": {
+                          "type": "string",
+                          "description": "The street address or other designation of a location.",
+                    },
+                    "addr2": {
+                        "type": "string",
+                        "description": "The street address or other designation of a location.",
+                        },
+                  },
+                "required": ["addr1", "addr2"],
+                },
+            },
+        ],
         previous_response_id=previous_response_id,
         instructions = instructions,
-        input=prompt,
+        input=inputs,
         reasoning ={
             "effort": "medium",
             "summary": "auto"
@@ -863,18 +899,28 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel, uploaded_ima
         using_groq = False
         reasoning = False
         prompt = prompt.strip()
-        if mode == 'Advanced': # and isBoss:
+        need_turn = True
+        responses = []
+        inputs = []
+        prev_id = None
+        if mode == "Advanced":
+            if len(past):
+                (prev_id, container_id) = past.pop()
+                past = []
+        while mode == 'Advanced' and need_turn:
+            need_turn = False
             if len(past) == 0:
-                prev_id = None
                 container_id = create_openai_container('My Container')
+                inputs.append(
+                    {"role": "user", "content": f"{prompt}"}
+                    )
             else:
-                (prev_id, container_id) = past[-1]
-                past.pop()
-                response = ''
+                (prev_id, container_id) = past.pop()
                 for item in past:
                     response += item
             try:
-                result = get_response(prompt, prev_id, container_id)
+                result = get_response(inputs, prev_id, container_id, uploaded_image_file)
+                uploaded_image_file = ''
             except:
                 return [[], "Sorry, there was an error getting the AI response",
                        prompt, gptModel, uploaded_image_file, plot, image_out, file_out]
@@ -882,6 +928,7 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel, uploaded_ima
             image_done = False
             ann_files = [] # (container_id, file_id, filename)
             code_files = [] # (container_id, file_id, filename)
+            text = '???  AI returned no text for this query'
             for output in result.output:
                 if output.type == 'message':
                     for content in output.content:
@@ -919,15 +966,53 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel, uploaded_ima
                             code_files.append((cont_id, file_id, file_name))
                     except:
                         pass
+                elif output.type == 'function_call':
+                    if output.name == 'get_distance':
+                        args = json.loads(output.arguments)
+                        distance = get_distance(args['addr1'], args['addr2'])
+                        inputs.append({
+                            "type": "function_call_output",
+                            "call_id": f"{output.call_id}",
+                            "output": f"{float(distance):.2f}",
+                            } )
+                        need_turn = True
+                        continue
+                elif output.type == 'image_generation_call':
+                    if result and not image_done:
+                        image_done = True;
+                        image_data = base64.b64decode(result)
+                        fpath = dataDir + user_window + '.png'
+                        with open(fpath,'wb') as fp:
+                            fp.write(image_data)
+                        image_out = gr.Image(visible=True, value=fpath)
             do_file_download = False
+            ext = ''
+            backup_image = None
             if len(ann_files) > 0:
                 (cont_id, file_id, file_name) = ann_files[-1]
+                ext = file_name.split('.')[-1].casefold()
                 do_file_download = True
             elif len(code_files) > 0:
-                (cont_id, file_id, file_name) = code_files[0]
-                do_file_download = True
+                for i in range(len(code_files)):
+                    (cont_id, file_id, file_name) = code_files[i]
+                    if file_name.casefold().find('access') >= 0:
+                        continue
+                    ext = file_name.split('.')[-1].casefold()
+                    if ext == 'png':
+                        if not image_done:
+                            backup_image = code_files[i]
+                    else:
+                        do_file_download = True
+                        break
+            if not do_file_download and not image_done and backup_image:
+                (cont_id, file_id, file_name) = backup_image
+                fpath = dataDir + user_window + '.png'
+                image_data = get_openai_file(file_id, cont_id).content
+                with open(fpath,'wb') as fp:
+                    fp.write(image_data)
+                image_out = gr.Image(visible=True, value=fpath)
+
             if do_file_download:
-                ext = file_name.split('.')[-1]
                 fpath = dataDir + user_window + '.' + ext
                 try:
                     data = get_openai_file(file_id, cont_id).content  
@@ -938,6 +1023,10 @@ def chat(prompt, user_window, pwd_window, past, response, gptModel, uploaded_ima
                 except:
                     text += f'\nUnable to load code-generated file: {file_name}'
                 # text += '\nIf a download link is given above, ignore it. Use the button below'
+            if need_turn:
+                # past.append(md(prompt))
+                past.append((result.id, container_id))
+                continue
             out_text = "\n".join(line for line in text.splitlines() if 
                                  'download' not in line.casefold())
             res = md("\n\n***YOU***: " + prompt + "\n\n***GPT***: " + out_text)
@@ -1289,10 +1378,10 @@ def show_help():
              can edit what's to be spoken.  Except:  In a chat conversation, spoken dialog will only include
              the latest prompt/response ("YOU:/GPT:") sequence.'''
     return str(txt).replace('```', ' ').replace('  ', '&nbsp;&nbsp;').replace('  ', '&nbsp;&nbsp;').replace('  ', '&nbsp;&nbsp;').replace('\n','<br>')
-def upload_image(prompt, user, password):
+def upload_image(prompt, user, password, mode):
     if not (user in unames and password == pwdList[unames.index(user)]):
         return [gr.Image(visible=False, interactive=True), "Incorrect user name and/or password"]
-    if len(prompt) < 3:
+    if len(prompt) < 3 and mode != 'Advanced':
         return [gr.Image(visible=False, interactive=True), "You must provide prompt/instructions (what to do with the image)"]
     return [gr.Image(visible=True, interactive=True), '']
 
@@ -1518,7 +1607,8 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
     button_do_image.click(fn=make_image, inputs=[prompt_window,user_window, password],outputs=[image_window, output_window])
     image_window.change(fn=delete_image, inputs=[user])
     help_button.click(fn=show_help, outputs=output_window)
-    button_get_image.click(fn=upload_image,inputs = [prompt_window, user, password], outputs = [image_window2, output_window])
+    button_get_image.click(fn=upload_image,inputs = [prompt_window, user, password, mode],
+                          outputs = [image_window2, output_window])
     image_window2.upload(fn=load_image, inputs=[image_window2, user], outputs=[uploaded_image_file, output_window])
     mode.change(fn=mode_change, inputs=mode,outputs=news_period)
     pwd_window.blur(updatePassword, inputs = pwd_window, outputs = [password, pwd_window, mode])