How can we show all previous threads in streamlit ui

[anuroop18]

I want all my previous chat threads to appear in sidebar in streamlit ui (Like in chatgpt). Because copying thread_id and pasting as url query string is not very useful for me.

Will i have to store each thread_id in a json file once it is created? Or is there any better approach?

[anuroop18]

I was able to achieve this using following code:

import json
from datetime import datetime
from pathlib import Path

# Define and create the data directory
PROJECT_ROOT = Path(__file__).parent.parent
DATA_DIR = PROJECT_ROOT / "data"
DATA_DIR.mkdir(exist_ok=True)
THREAD_HISTORY_FILE = DATA_DIR / "thread_history.json"

def load_thread_history() -> dict:
    try:
        with open(THREAD_HISTORY_FILE, "r") as f:
            return json.load(f)
    except FileNotFoundError:
        return {}

def save_thread_history(threads: dict) -> None:
    with open(THREAD_HISTORY_FILE, "w") as f:
        json.dump(threads, f)

def update_thread_access(thread_id: str) -> None:
    threads = load_thread_history()
    threads[thread_id] = datetime.now().isoformat()
    save_thread_history(threads)

Then we can use that function like following:

# Generate new message if the user provided new input
    if user_input := st.chat_input():
        messages.append(ChatMessage(type="human", content=user_input))
        st.chat_message("human").write(user_input)
        # Update thread access when sending a new message
        update_thread_access(st.session_state.thread_id)

And we can display the threads like following:

# Config options
  with st.sidebar:
      st.divider()
      st.subheader("Previous Conversations")
      # Load and sort threads by last access time
      threads = load_thread_history()
      sorted_threads = sorted(threads.items(), key=lambda x: x[1], reverse=True)

      for thread_id, last_accessed in sorted_threads:
          last_accessed_dt = datetime.fromisoformat(last_accessed)
          display_time = last_accessed_dt.strftime("%Y-%m-%d %H:%M")
          # Create clickable link with the current URL base and target="_self"
          thread_url = f"?thread_id={thread_id}"
          st.markdown(f'<a href="{thread_url}" target="_self">Thread: {thread_id[:8]}... ({display_time})</a>', unsafe_allow_html=True)

Is there a better approach??
How do i get nice summary of that thread like in chatgpt??

[anuroop18]

And these threads are not preserved over restarts of server.. how to achieve that?

[JoshuaC215]

Hi Anuroop, nice! Your approach makes sense. For a quick summary, you would want to make a separate call to an LLM asking it to generate a title based on the provided conversation. Here's an example prompt I wrote in a previous project:

https://github.com/streamlit/llm-examples/blob/llm-eval/llm-eval/common_ui.py#L168-L190

Once you have the title, you could add that to the thread dictionary as a new key.

One thing to note - with your approach it will work locally but:

• On a cloud deployment, the JSON file might not be persisted unless you put it on a durable volume. Alternatively you could store it in a database that you provide (this is not possible in Streamlit Cloud, you'd need to set up an external database)
• Since you used a global file, if multiple users are interacting with the app they would all see the same message history. To solve this you would need some kind of user identity (like a browser cookie or login) and then also include that in the conversation history.

The example project I linked above had many of these features although I didn't use a cookie for login. It's a somewhat complex app and I haven't tried to integrate it with agent-service-toolkit. I don't really have capacity to work on that but maybe it will be a good reference for you or anyone else.

You can view the app here: https://llm-eval-demo.streamlit.app/about

[anuroop18]

I’ll review the example project you shared to integrate similar features into my app. I really appreciate your help. Thanks a lot.. 😊