updated

Dockerfile

@@ -7,12 +7,15 @@ COPY ./requirements.txt /app/requirements.txt
 RUN pip3 install --no-cache-dir --upgrade pip
 RUN pip3 install --no-cache-dir wheel setuptools build
 RUN pip3 install --no-cache-dir --use-pep517 -r /app/requirements.txt
-
+ 
 # User
 RUN useradd -m -u 1000 user
 USER user
 ENV HOME /home/user
 ENV PATH $HOME/.local/bin:$PATH
+ENV TIKTOKEN_CACHE_DIR $HOME/.cache/tiktoken
+
+RUN mkdir -p $HOME/.cache/tiktoken
 
 WORKDIR $HOME
 RUN mkdir app

agent.py

@@ -10,7 +10,7 @@ from omegaconf import OmegaConf
 from vectara_agentic.agent import Agent
 from vectara_agentic.tools import ToolsFactory, VectaraToolFactory
 from vectara_agentic.agent_config import AgentConfig
-from vectara_agentic.sub_query_workflow import SequentialSubQuestionsWorkflow
+from vectara_agentic.types import ModelProvider, AgentType
 
 from dotenv import load_dotenv
 load_dotenv(override=True)
@@ -33,7 +33,7 @@ tickers = {
     "STT": "State Street",
     "BK": "Bank of New York Mellon",
 }
-years = range(2015, 2025)
+years = list(range(2015, 2025))
 initial_prompt = "How can I help you today?"
 
 
@@ -102,47 +102,36 @@ class AgentTools:
 
     def get_tools(self):
         class QueryTranscriptsArgs(BaseModel):
-            query: str = Field(..., description="The user query, always in the form of a question", examples=["what are the risks reported?", "who are the competitors?"])
+            ticker: str = Field(description="The ticker symbol for the company", examples=list(tickers.keys()), default=None)
             year: int | str = Field(
                 default=None,
                 description=f"The year this query relates to. An integer between {min(years)} and {max(years)} or a string specifying a condition on the year",
                 examples=[2020, '>2021', '<2023', '>=2021', '<=2023', '[2021, 2023]', '[2021, 2023)']
             )
-            ticker: str = Field(..., description=f"The company ticker this query relates to. Must be a valid ticket symbol from the list {list(tickers.keys())}.")
 
-        vec_factory = VectaraToolFactory(vectara_api_key=self.cfg.api_key,
-                                        vectara_corpus_key=self.cfg.corpus_key)
         summarizer = 'vectara-summary-table-md-query-ext-jan-2025-gpt-4o'
-        ask_transcripts = vec_factory.create_rag_tool(
+        ask_transcripts = self.vec_factory.create_rag_tool(
             tool_name = "ask_transcripts",
             tool_description = """
-            Given a company name and year, responds to a user question about the company, based on analyst call transcripts about the company's financial reports for that year.
-            You can ask this tool any question about the company including risks, opportunities, financial performance, competitors and more.
+            Answer questions about a company (using its ticker) including risks, opportunities, financial performance, competitors and more, based on earnings calls transcripts.
             """,
             tool_args_schema = QueryTranscriptsArgs,
-            reranker = "multilingual_reranker_v1", rerank_k = 100, rerank_cutoff = 0.2,
-            n_sentences_before = 2, n_sentences_after = 4, lambda_val = 0.005,
+            reranker = "multilingual_reranker_v1", rerank_k = 100, rerank_cutoff = 0.3,
+            n_sentences_before = 2, n_sentences_after = 2, lambda_val = 0.005,
             summary_num_results = 15,
             vectara_summarizer = summarizer,
+            max_tokens = 4096, max_response_chars = 8192,
             include_citations = True,
+            save_history = True,
             verbose = False,
         )
 
-        class SearchTranscriptsArgs(BaseModel):
-            query: str = Field(..., description="The user query, always in the form of a question", examples=["what are the risks reported?", "who are the competitors?"])
-            top_k: int = Field(..., description="The number of results to return.")
-            year: int | str = Field(
-                default=None,
-                description=f"The year this query relates to. An integer between {min(years)} and {max(years)} or a string specifying a condition on the year",
-                examples=[2020, '>2021', '<2023', '>=2021', '<=2023', '[2021, 2023]', '[2021, 2023)']
-            )
-            ticker: str = Field(..., description=f"The company ticker this query relates to. Must be a valid ticket symbol from the list {list(tickers.keys())}.")
-        search_transcripts = vec_factory.create_search_tool(
+        search_transcripts = self.vec_factory.create_search_tool(
             tool_name = "search_transcripts",
             tool_description = """
-            Given a company name and year, and a user query, retrieves relevant documents about the company.
+            retrieves relevant earning call transcripts about a company (using its ticker).
             """,
-            tool_args_schema = SearchTranscriptsArgs,
+            tool_args_schema = QueryTranscriptsArgs,
             reranker = "multilingual_reranker_v1", rerank_k = 100,
             lambda_val = 0.005,
             verbose=False
@@ -156,13 +145,14 @@ class AgentTools:
                         get_valid_years,
                         fmp_income_statement,
                     ]
-                ] +
+                ] + 
                 [ask_transcripts, search_transcripts]
         )
 
 def initialize_agent(_cfg, agent_progress_callback=None) -> Agent:
     financial_bot_instructions = """
-    - You are a helpful financial assistant, with expertise in financial reporting, in conversation with a user. 
+    - You are a helpful financial assistant, with expertise in financial reporting, in conversation with a user.
+    - Never base your on general industry knowledge, only use information from tool calls.
     - Use the 'fmp_income_statement' tool (with the company ticker and year) to obtain financial data.
     - Always check the 'get_company_info' and 'get_valid_years' tools to validate company and year are valid.
     - Use the 'ask_transcripts' tool to answer most questions about the company's financial performance, risks, opportunities, strategy, competitors, and more.
@@ -177,19 +167,35 @@ def initialize_agent(_cfg, agent_progress_callback=None) -> Agent:
     def query_logging(query: str, response: str):
         print(f"Logging query={query}, response={response}")
 
-    agent_config = AgentConfig()
+    agent_config = AgentConfig(
+        agent_type = os.getenv("VECTARA_AGENTIC_AGENT_TYPE", AgentType.OPENAI.value),
+        main_llm_provider = os.getenv("VECTARA_AGENTIC_MAIN_LLM_PROVIDER", ModelProvider.OPENAI.value),
+        main_llm_model_name = os.getenv("VECTARA_AGENTIC_MAIN_MODEL_NAME", ""),
+        tool_llm_provider = os.getenv("VECTARA_AGENTIC_TOOL_LLM_PROVIDER", ModelProvider.OPENAI.value),
+        tool_llm_model_name = os.getenv("VECTARA_AGENTIC_TOOL_MODEL_NAME", ""),
+        observer = os.getenv("VECTARA_AGENTIC_OBSERVER_TYPE", "NO_OBSERVER")
+    )
+    fallback_agent_config = AgentConfig(
+        agent_type = os.getenv("VECTARA_AGENTIC_FALLBACK_AGENT_TYPE", AgentType.OPENAI.value),
+        main_llm_provider = os.getenv("VECTARA_AGENTIC_FALLBACK_MAIN_LLM_PROVIDER", ModelProvider.OPENAI.value),
+        main_llm_model_name = os.getenv("VECTARA_AGENTIC_FALLBACK_MAIN_MODEL_NAME", ""),
+        tool_llm_provider = os.getenv("VECTARA_AGENTIC_FALLBACK_TOOL_LLM_PROVIDER", ModelProvider.OPENAI.value),
+        tool_llm_model_name = os.getenv("VECTARA_AGENTIC_FALLBACK_TOOL_MODEL_NAME", ""),
+        observer = os.getenv("VECTARA_AGENTIC_OBSERVER_TYPE", "NO_OBSERVER")
+    )
 
     agent = Agent(
+        agent_config=agent_config,
+        fallback_agent_config=fallback_agent_config,
         tools=AgentTools(_cfg, agent_config).get_tools(),
         topic="Financial data, annual reports and 10-K filings",
         custom_instructions=financial_bot_instructions,
         agent_progress_callback=agent_progress_callback,
         query_logging_callback=query_logging,
         verbose=True,
-        workflow_cls=SequentialSubQuestionsWorkflow,
     )
 
-    agent.report()
+    agent.report(detailed=False)
     return agent
 
 def get_agent_config() -> OmegaConf:

requirements.txt

@@ -1,9 +1,9 @@
 omegaconf==2.3.0
 python-dotenv==1.0.1
-streamlit==1.43.2
+streamlit==1.45.0
 streamlit_feedback==0.1.3
 uuid==1.30
 langdetect==1.0.9
 langcodes==3.4.0
-vectara-agentic==0.2.11
+vectara-agentic==0.2.15
 torch==2.6.0

st_app.py

@@ -14,22 +14,12 @@ from agent import initialize_agent, get_agent_config
 
 initial_prompt = "How can I help you today?"
 
-# def pil_to_base64(img):
-#     buffered = BytesIO()
-#     img.save(buffered, format="PNG")
-#     return base64.b64encode(buffered.getvalue()).decode()
-
-
 def format_log_msg(log_msg: str):
     max_log_msg_size = 500
     return log_msg if len(log_msg) <= max_log_msg_size else log_msg[:max_log_msg_size]+'...'
 
 def agent_progress_callback(status_type: AgentStatusType, msg: str):
     output = f'<span style="color:blue;">{status_type.value}</span>: {msg}'
-    if "log_messages" not in st.session_state:
-        st.session_state.log_messages = [output]
-    else:
-        st.session_state.log_messages.append(output)
     st.session_state.log_messages.append(output)
     if 'status' in st.session_state:
         latest_message = ''
@@ -85,6 +75,7 @@ async def launch_bot():
         else:
             st.session_state.agent.clear_memory()
 
+
     if 'cfg' not in st.session_state:
         cfg = get_agent_config()
         st.session_state.cfg = cfg
@@ -154,7 +145,7 @@ async def launch_bot():
     if st.session_state.prompt:
         with st.chat_message("assistant", avatar='🤖'):
             st.session_state.status = st.status('Processing...', expanded=False)
-            response = st.session_state.agent.chat(st.session_state.prompt)
+            response = await st.session_state.agent.achat(st.session_state.prompt)
             res = escape_dollars_outside_latex(response.response)
 
             #from vectara_agentic.sub_query_workflow import SequentialSubQuestionsWorkflow