Use custom chat interface

app.py

@@ -63,6 +63,11 @@ def run_workflow(input, history, thread_id):
 
     print(f"Using thread_id: {thread_id}")
 
+    # Display the user input in the chatbot
+    history.append(gr.ChatMessage(role="user", content=input))
+    # Return the message history and empty lists for emails and citations texboxes
+    yield history, [], []
+
     # Asynchronously stream graph steps for a single input
     # https://langchain-ai.lang.chat/langgraph/reference/graphs/#langgraph.graph.state.CompiledStateGraph
     for step in graph.stream(
@@ -98,6 +103,7 @@ def run_workflow(input, history, thread_id):
                         )
                     )
             if chunk_messages.content:
+                # Display response made instead of or in addition to a tool call
                 history.append(
                     gr.ChatMessage(role="assistant", content=chunk_messages.content)
                 )
@@ -214,6 +220,20 @@ with gr.Blocks(
     css=css,
 ) as demo:
 
+    # -----------------
+    # Data availability
+    # -----------------
+
+    def is_data_present():
+        """Check if the database directory is present"""
+
+        return os.path.isdir(db_dir)
+
+    def is_data_missing():
+        """Check if the database directory is missing"""
+
+        return not os.path.isdir(db_dir)
+
     # -----------------
     # Define components
     # -----------------
@@ -233,7 +253,7 @@ with gr.Blocks(
     downloading = gr.Textbox(
         lines=1,
         label="Downloading Data, Please Wait",
-        visible=False,
+        visible=is_data_missing(),
         render=False,
     )
     extracting = gr.Textbox(
@@ -268,6 +288,29 @@ with gr.Blocks(
         show_copy_all_button=True,
         render=False,
     )
+    # Modified from gradio/chat_interface.py
+    input = gr.Textbox(
+        show_label=False,
+        label="Message",
+        placeholder="Type a message...",
+        scale=7,
+        autofocus=True,
+        submit_btn=True,
+        render=False,
+    )
+    emails_textbox = gr.Textbox(
+        label="Retrieved Emails",
+        info="Tip: Look for 'Tool Call' and 'Next Email' separators. Quoted lines (starting with '>') are removed before indexing.",
+        lines=10,
+        visible=False,
+        render=False,
+    )
+    citations_textbox = gr.Textbox(
+        label="Citations",
+        lines=2,
+        visible=False,
+        render=False,
+    )
 
     # ------------
     # Set up state
@@ -314,7 +357,7 @@ with gr.Blocks(
         if compute_mode.startswith("edge"):
             status_text = f"""
             📍 Now in **edge** mode, using ZeroGPU hardware<br>
-            ⌛ Response time is ca. 2-3 minutes<br>
+            ⌛ Response time is around 2 minutes<br>
             ✨ Embeddings: [Nomic](https://huggingface.co/nomic-ai/nomic-embed-text-v1.5); LLM: [{model_id}](https://huggingface.co/{model_id})<br>
             🏠 See the project's [GitHub repository](https://github.com/jedick/R-help-chat)
             """
@@ -339,31 +382,21 @@ with gr.Blocks(
         return info_text
 
     with gr.Row():
-        # Left column: Intro, Compute, Chat, Emails
+        # Left column: Intro, Compute, Chat
         with gr.Column(scale=2):
             with gr.Row(elem_classes=["row-container"]):
                 with gr.Column(scale=2):
                     intro = gr.Markdown(get_intro_text())
                 with gr.Column(scale=1):
                     compute_mode.render()
-            chat_interface = gr.ChatInterface(
-                to_workflow,
-                chatbot=chatbot,
-                type="messages",
-                additional_inputs=[thread_id],
-                additional_outputs=[retrieved_emails, citations_text],
-                api_name=False,
-            )
+            with gr.Group(visible=is_data_present()) as chat_interface:
+                chatbot.render()
+                input.render()
+            # Render textboxes for data loading progress
             downloading.render()
             extracting.render()
             data_error.render()
-            emails_textbox = gr.Textbox(
-                label="Retrieved Emails",
-                lines=10,
-                visible=False,
-                info="Tip: Look for 'Tool Call' and 'Next Email' separators. Quoted lines (starting with '>') are removed before indexing.",
-            )
-        # Right column: Info, Examples, Citations
+        # Right column: Info, Examples
         with gr.Column(scale=1):
             status = gr.Markdown(get_status_text(compute_mode.value))
             with gr.Accordion("ℹ️ More Info", open=False):
@@ -380,7 +413,7 @@ with gr.Blocks(
                 ]
                 gr.Examples(
                     examples=[[q] for q in example_questions],
-                    inputs=[chat_interface.textbox],
+                    inputs=[input],
                     label="Click an example to fill the message box",
                     elem_id="example-questions",
                 )
@@ -390,7 +423,7 @@ with gr.Blocks(
                 ]
                 gr.Examples(
                     examples=[[q] for q in multi_tool_questions],
-                    inputs=[chat_interface.textbox],
+                    inputs=[input],
                     label="Multiple retrievals (cloud mode)",
                     elem_id="example-questions",
                 )
@@ -400,11 +433,17 @@ with gr.Blocks(
                 ]
                 gr.Examples(
                     examples=[[q] for q in multi_turn_questions],
-                    inputs=[chat_interface.textbox],
+                    inputs=[input],
                     label="Asking follow-up questions",
                     elem_id="example-questions",
                 )
-            citations_textbox = gr.Textbox(label="Citations", lines=2, visible=False)
+
+    # Bottom row: retrieved emails and citations
+    with gr.Row():
+        with gr.Column(scale=2):
+            emails_textbox.render()
+        with gr.Column(scale=1):
+            citations_textbox.render()
 
     # -------------
     # App functions
@@ -434,23 +473,23 @@ with gr.Blocks(
         """Return updated visibility state for a component"""
         return gr.update(visible=visible)
 
-    def update_emails(retrieved_emails, emails_textbox):
-        if retrieved_emails is None:
-            # This keeps the content of the textbox when the answer step occurs
-            return emails_textbox, change_visibility(True)
-        elif not retrieved_emails == []:
-            # This adds the retrieved emails to the textbox
-            return retrieved_emails, change_visibility(True)
-        else:
-            # This blanks out the textbox when a new chat is started
-            return "", change_visibility(False)
-
-    def update_citations(citations):
-        if citations == []:
-            # Blank out and hide the citations textbox when new input is submitted
+    def update_textbox(content, textbox):
+        if content is None:
+            # Keep the content of the textbox unchanged
+            return textbox, change_visibility(True)
+        elif content == []:
+            # Blank out the textbox
             return "", change_visibility(False)
         else:
-            return citations, change_visibility(True)
+            # Display the content in the textbox
+            return content, change_visibility(True)
+
+    #    def update_citations(citations):
+    #        if citations == []:
+    #            # Blank out and hide the citations textbox when new input is submitted
+    #            return "", change_visibility(False)
+    #        else:
+    #            return citations, change_visibility(True)
 
     # --------------
     # Event handlers
@@ -494,9 +533,17 @@ with gr.Blocks(
         api_name=False,
     )
 
+    input.submit(
+        # Submit input to the chatbot
+        to_workflow,
+        [input, chatbot, thread_id],
+        [chatbot, retrieved_emails, citations_text],
+        api_name=False,
+    )
+
     retrieved_emails.change(
         # Update the emails textbox
-        update_emails,
+        update_textbox,
         [retrieved_emails, emails_textbox],
         [emails_textbox, emails_textbox],
         api_name=False,
@@ -504,8 +551,8 @@ with gr.Blocks(
 
     citations_text.change(
         # Update the citations textbox
-        update_citations,
-        [citations_text],
+        update_textbox,
+        [citations_text, citations_textbox],
         [citations_textbox, citations_textbox],
         api_name=False,
     )
@@ -514,16 +561,6 @@ with gr.Blocks(
     # Data loading
     # ------------
 
-    def is_data_present():
-        """Check if the database directory is present"""
-
-        return os.path.isdir(db_dir)
-
-    def is_data_missing():
-        """Check if the database directory is missing"""
-
-        return not os.path.isdir(db_dir)
-
     def download():
         """Download the db.zip file"""
 
@@ -555,30 +592,22 @@ with gr.Blocks(
         return None
 
     false = gr.State(False)
-    true = gr.State(True)
     need_data = gr.State()
     have_data = gr.State()
 
     # When app is launched, check if data is present, download it if necessary,
-    # hide chat interface during downloading, show downloading and extracting
-    # steps as textboxes, show error textbox if needed, restore chat interface,
-    # and update database info
+    # show extracting step as textbox, show error textbox if needed,
+    # update database info, and restore chat interface.
+    # nb. initial visibility of chat interface components and
+    # downloading textbox are set after instantiation of ChatInterface above
 
     # fmt: off
     demo.load(
-        is_data_missing, None, [need_data], api_name=False
-    ).then(
-        is_data_present, None, [have_data], api_name=False
-    ).then(
-        change_visibility, [have_data], [chatbot], api_name=False
-    ).then(
-        change_visibility, [have_data], [chat_interface.textbox], api_name=False
-    ).then(
-        change_visibility, [need_data], [downloading], api_name=False
-    ).then(
         download, None, [downloading], api_name=False
     ).then(
         change_visibility, [false], [downloading], api_name=False
+    ).then(
+        is_data_missing, None, [need_data], api_name=False
     ).then(
         change_visibility, [need_data], [extracting], api_name=False
     ).then(
@@ -586,17 +615,15 @@ with gr.Blocks(
     ).then(
         change_visibility, [false], [extracting], api_name=False
     ).then(
-        is_data_missing, None, [need_data], api_name=False
+        get_info_text, None, [info], api_name=False
     ).then(
         is_data_present, None, [have_data], api_name=False
     ).then(
-        change_visibility, [have_data], [chatbot], api_name=False
+        change_visibility, [have_data], [chat_interface], api_name=False
     ).then(
-        change_visibility, [have_data], [chat_interface.textbox], api_name=False
+        is_data_missing, None, [need_data], api_name=False
     ).then(
         change_visibility, [need_data], [data_error], api_name=False
-    ).then(
-        get_info_text, None, [info], api_name=False
     )
     # fmt: on
 

prompts.py

@@ -17,15 +17,14 @@ def retrieve_prompt(compute_mode):
         f"Today Date: {date.today()}. "
         "You are a helpful RAG chatbot designed to answer questions about R programming based on the R-help mailing list. "
         "Do not ask the user for more information, but retrieve emails from the R-help mailing list archives. "
+        # gpt-4o-mini says last two months aren't available with this: Emails from from {start} to {end} are available for retrieval.
         f"The emails available for retrieval are from {start} to {end}. "
-        "Write a search query based on the user'r question, but do not answer the question just yet. "
+        "Write a search query based on the user's question, but do not answer the question just yet. "
         "For questions about differences or comparison between X and Y, retrieve emails about X and Y. "
-        "Use the 'months' argument to search for months. "
+        "If the user's question is about years, use retrieve_emails(search_query=, start_year=, end_year=) (this month is this year). "
+        "Example: to retrieve emails about R from a month in any year use retrieve_emails(search_query='R', months=<month>). "
         # This confuses gpt-4o-mini (empty search_query - token problem?)
-        # "Use 3-letter month abbreviations (Jan, Feb, Mar, Apr, May, Jun, Jul, Aug, Sep, Oct, Nov, Dec). "
-        "Example: retrieve emails about R in <month> using retrieve_emails(search_query='R', months=<month>). "
-        "Try to infer years from the user's question (last month or this month is this year). "
-        "If you can infer years, use retrieve_emails(search_query=, start_year=, end_year=), otherwise retrieve_emails(search_query=). "
+        "Use 3-letter month abbreviations: Jan for January, Jul for July. "
         "If you decide not to retrieve emails, tell the user why and suggest how to improve their question to chat with the R-help mailing list. "
     )
     # A sanity check that we don't have unassigned variables
@@ -52,9 +51,7 @@ def answer_prompt(with_tools=True):
         "Respond with 300 words maximum and 30 lines of code maximum and include any relevant URLs from the retrieved emails. "
     )
     if with_tools:
-        answer_prompt += (
-            "Use answer_with_citations to provide the answer and all citations used. "
-        )
+        answer_prompt += "Use answer_with_citations to provide the complete answer and all citations used. "
     matches = re.findall(r"\{.*?\}", "".join(answer_prompt))
     if matches:
         raise ValueError(f"Unassigned variables in prompt: {' '.join(matches)}")