Agentic streamlit chatbot

examples/README.md

@@ -125,6 +125,11 @@ it is useful when you want to quickly do ranking experiments without rewriting a
 * [MLCon](https://mlconference.ai/machine-learning-advanced-development/adaptive-incontext-learning/)
 * [data science connect COLLIDE](https://datasciconnect.com/events/collide/agenda/)
 
+### Agentic Chatbot using Vespa
+
+[![logo](/assets/vespa-logomark-tiny.png) agentic-streamlit-chatbot](agentic-streamlit-chatbot) This sample Streamlit application demonstrates how to use [LangGraph](https://www.langchain.com/langgraph) agentic framework to develop an E-commerce chatbot using Vespa as a retrieval tool.
+
+
 For any questions, please register at the Vespa Slack and discuss in the general channel.
 
 ----

examples/agentic-streamlit-chatbot/README.md

@@ -0,0 +1,66 @@
+# Prerequisites
+
+
+- You need a python virtual environment. For this demo, `Python 3.13.1` was used, but any Python environment 3.11+ should work.
+- You will need [Vespa CLI](https://docs.vespa.ai/en/vespa-cli.html) that you can deploy on MacOS with `brew install vespa-cli`
+- Libraries dependencies can be installed with: `pip install -R requirements.txt`
+- Sign-up with [Tavily](https://tavily.com/) and Get an API key.
+- Spin-up a Vespa Cloud [Trial](https://vespa.ai/free-trial) account:
+  - Login to the account you just created and create a tenant at [console.vespa-cloud.com](https://console.vespa-cloud.com/).
+  - Save the tenant name.
+- A Valid OpenAI API key. Note that you have the option to use any other LLM, which may support Langgraph tools binding.
+- Git clone the repo `https://github.com/vespa-engine/system-test.git`
+- The ecommerce_hybrid_search app will be deployed. For more information about the app, please review the [README.md](https://github.com/vespa-engine/system-test/blob/master/tests/performance/ecommerce_hybrid_search/dataprep/README.md). You do not have to follow the data prep steps there. Follow the instructions below instead.
+- Uncompress the data file: `zstd -d data/vespa_update-96k.json.zst`
+
+
+
+# Deploy the Ecommerce Vespa Application
+
+
+- In the system-test repo you just cloned, navigate to `tests/performance/ecommerce_hybrid_search/app`
+- Choose a name for your app. For example `ecommercebot`
+- Follow instructions in the [**Getting Started**](https://cloud.vespa.ai/en/getting-started) document. Please note the following as you go through the documentation:
+  - You will need your tenant name you created previously.
+  - When adding the public certificates with `vespa auth cert`, it will give you the absolute path of the certificate and the private key. Please note them down.
+  - To feed the application, return to the original directory and run:
+  ```
+  vespa feed data/vespa_update-96k.json
+  ```
+- You can test the following query from the Vespa CLI:
+  ```
+  vespa query "select id, category, title, price  from sources * where default contains 'screwdriver'"
+  ```
+ - You will need the URL of your Vespa application. Run the following command:
+  ```
+  vespa status
+  ```
+  This should return you an output like:
+  ```
+  Container container at https://xxxxx.yyyyy.z.vespa-app.cloud/ is ready
+  ```
+  Note down the URL.
+
+  # Configure and Launch your Streamlit Application
+
+  A template for `secrets.toml` file to store streamlit secrets has been provided. Please create a subdirectory `.streamlit` and copy the template there. 
+
+  Update all the fields with all the information collected previously and save the file as `secrets.toml`
+
+  Launch your streamlit application:
+  ```
+  streamlit run streamlit_vespa_app.py
+  ```
+  # Testing the Application
+
+  You can try a mix of questions like:
+
+  `What is the weather in Toronto ?`
+
+  Followed by:
+
+  `I'm looking for a screwdriver`
+
+  And then:
+
+  `Which one do you recommend to fix a watch?`

examples/agentic-streamlit-chatbot/requirements.txt

@@ -0,0 +1,6 @@
+pyvespa
+langchain_openai 
+langchain_core 
+langchain_community
+langgraph
+streamlit

examples/agentic-streamlit-chatbot/streamlit_vespa_app.py

@@ -0,0 +1,141 @@
+
+import streamlit as st
+import getpass
+import os
+
+from vespa.application import Vespa
+from vespa.io import VespaResponse, VespaQueryResponse
+import json
+
+from langchain_community.tools.tavily_search import TavilySearchResults
+
+from langchain_openai import ChatOpenAI
+
+from langgraph.graph import MessagesState
+from langchain_core.messages import AIMessage, HumanMessage, SystemMessage
+
+from langgraph.graph import START, StateGraph
+from langgraph.prebuilt import tools_condition, ToolNode
+
+from PIL import Image
+
+st.title(":shopping_trolley: Vespa.ai Shopping Assistant")
+
+# Load an image (ensure the file is in the correct path)
+icon = Image.open("Vespa-logo-dark-RGB.png")
+
+# Display the image in the sidebar
+st.sidebar.image(icon, width=500)
+
+# Fetch secrets
+OPENAI_API_KEY = st.secrets["api_keys"]["llm"]
+TAVILY_API_KEY = st.secrets["api_keys"]["tavily"]
+VESPA_URL = st.secrets["vespa"]["url"]
+PUBLIC_CERT_PATH = st.secrets["vespa"]["public_cert_path"]
+PRIVATE_KEY_PATH = st.secrets["vespa"]["private_key_path"]
+
+#if not os.environ.get("OPENAI_API_KEY"):
+os.environ["OPENAI_API_KEY"] = OPENAI_API_KEY
+#if not os.environ.get("TAVILY_API_KEY"):
+os.environ["TAVILY_API_KEY"] = TAVILY_API_KEY
+
+
+def VespaRetriever(UserQuery: str) -> str:
+    """Retrieves all items for sale matching the user query.
+
+    Args:
+        UserQuery: User Query for items they are looking for.
+    """
+
+    vespa_app = Vespa(url=VESPA_URL,
+                  cert=PUBLIC_CERT_PATH,
+                  key=PRIVATE_KEY_PATH)
+
+    with vespa_app.syncio(connections=1) as session:
+        query = UserQuery
+        response: VespaQueryResponse = session.query(
+            yql="select id, category, title, average_rating, price from sources * where userQuery()",
+            query=query,
+            hits=5,
+            ranking="hybrid",
+        )
+    assert response.is_successful()
+
+    # Extract only the 'fields' content from each entry
+    filtered_data = [hit["fields"] for hit in response.hits]
+
+    # Convert to a JSON string
+    json_string = json.dumps(filtered_data, indent=1)
+
+    return json_string
+
+TavilySearch = TavilySearchResults(max_results=2)
+
+tools = [TavilySearch, VespaRetriever]
+
+llm = ChatOpenAI(model="gpt-4o")
+llm_with_tools = llm.bind_tools(tools)
+
+# System message
+sys_msg = SystemMessage(content="You are a helpful sales assistant willing to answer any user questions about items to sell. You will try your best to provide all the information regarding an item for sale to a customer.")
+
+# Node
+def assistant(state: MessagesState):
+   return {"messages": [llm_with_tools.invoke([sys_msg] + state["messages"])]}
+
+# Graph
+builder = StateGraph(MessagesState)
+
+# Define nodes: these do the work
+builder.add_node("assistant", assistant)
+builder.add_node("tools", ToolNode(tools))
+
+# Define edges: these determine how the control flow moves
+builder.add_edge(START, "assistant")
+builder.add_conditional_edges(
+    "assistant",
+    # If the latest message (result) from assistant is a tool call -> tools_condition routes to tools
+    # If the latest message (result) from assistant is a not a tool call -> tools_condition routes to END
+    tools_condition,
+)
+builder.add_edge("tools", "assistant")
+
+react_graph = builder.compile()
+
+#Initialize the chat messages history
+if "messages" not in st.session_state.keys():
+    st.session_state.messages = [
+       {"role": "assistant", "content": "Hello, I'm your Vespa Shopping Assistant using an agentic architecture based on LangGraph. How can I assist you today ?"}
+    ]
+
+#Prompt the user input and save
+if prompt := st.chat_input():
+    st.session_state.messages.append({"role": "user", "content": prompt})
+
+#Display the existing chat messages
+for message in st.session_state.messages:
+    with st.chat_message(message["role"]):
+       st.write(message["content"])
+
+#If last message is not from the assistant, we need to generate a new response
+if st.session_state.messages[-1]["role"] != "assistant":
+   question = st.session_state.messages[-1]["content"].replace('?','')
+
+   message_list = []
+   messages = react_graph.invoke({"messages": st.session_state.messages}, stream_mode="values")
+   print(messages)
+
+   for m in messages['messages']:
+       message_list.append(m)
+
+   print(message_list)
+
+   response_text = next(
+    (msg.content for msg in reversed(messages['messages']) if isinstance(msg, AIMessage)),
+    None
+    )
+   with st.chat_message("assistant"):
+       st.write("Response: ", response_text)
+
+       # **Add the assistant response to session state**
+   st.session_state.messages.append({"role": "assistant", "content": response_text})

examples/agentic-streamlit-chatbot/template-secrets.toml

@@ -0,0 +1,9 @@
+# .streamlit/secrets.toml
+
+[vespa]
+url = "https://<host>.vespa-app.cloud/"
+public_cert_path = "PATH/data-plane-public-cert.pem"
+private_key_path = "PATH/data-plane-private-key.pem"
+[api_keys]
+llm = "sk-"
+tavily = "tvly-"