Update langchain-chroma to fix ChromaDB ValueError

app.py

@@ -72,7 +72,7 @@ def run_workflow(input, history, compute_mode, thread_id, session_hash):
                 print_exception=False,
             )
 
-    # Get graph for compute mode
+    # Get graph instance
     graph = graph_instances[compute_mode].get(session_hash)
     if graph is not None:
         print(f"Get {compute_mode} graph for session {session_hash}")
@@ -274,7 +274,7 @@ with gr.Blocks(
     )
 
     loading_data = gr.Textbox(
-        "Please wait for data loading to complete.",
+        "Please wait for the email database to be downloaded and extracted.",
         max_lines=0,
         label="Loading Data",
         visible=False,

prompts.py

@@ -61,7 +61,7 @@ def generate_prompt(chat_model, think=False, with_tools=False):
         "You must include inline citations (email senders and dates) in each part of your response. "
         "Only answer general questions about R if the answer is in the retrieved emails. "
         "Your response can include URLs, but make sure they are unchanged from the retrieved emails. "  # Qwen
-        "Respond with 300 words maximum and 30 lines of code maximum. "
+        "Respond with 500 words maximum and 50 lines of code maximum. "
     )
     if with_tools:
         prompt = (

requirements.txt

@@ -1,30 +1,42 @@
-langchain>=0.3.25,<1.0
-langchain-community>=0.3.24,<1.0
-langchain-chroma==0.2.4
-langchain-huggingface==0.3.0
-langgraph>=0.4.7,<0.6
-# Required by langchain-huggingface
-sentence-transformers==5.0.0
-# Required by Nomic embeddings
-einops==0.8.1
+# Pin torch and chroma versions
 torch==2.5.1
+chromadb==1.0.13
+
 # Stated requirements:
-#   SmolLM3: transformers>=4.53
 #   Gemma 3: transformers>=4.50
+#   Qwen3:   transformers>=4.51
+#   SmolLM3: transformers>=4.53
 # NOTE: Gemma 3 with transformers==4.54.0 gives:
 #   ValueError: Max cache length is not consistent across layers
 transformers==4.51.3
+
+# Langchain packages
+langchain==0.3.25
+langchain-community==0.3.24
+langchain-chroma==0.2.5
+langchain-huggingface==0.3.0
+langgraph==0.4.7
+
+# Required by langchain-huggingface
+sentence-transformers==5.0.0
+# Required by Nomic embeddings
+einops==0.8.1
+
 # Commented because we have local modifications
 #tool-calling-llm==0.1.2
 bm25s==0.2.12
 ragas==0.2.15
+
 # posthog<6.0.0 is temporary fix for ChromaDB telemetry error log messages
 # https://github.com/vanna-ai/vanna/issues/917
 posthog==5.4.0
+
 # Gradio for the web interface
 gradio==5.38.2
 spaces==0.37.1
+
 # For downloading data from S3
 boto3==1.39.14
+
 # Others
-python-dotenv
+python-dotenv==1.1.1

retriever.py

@@ -174,9 +174,6 @@ def BuildRetrieverDense(compute_mode: str, top_k=6):
         # Get top k documents
         search_kwargs={"k": top_k},
     )
-    # Fix for ValueError('Could not connect to tenant default_tenant. Are you sure it exists?')
-    # https://github.com/langchain-ai/langchain/issues/26884
-    chromadb.api.client.SharedSystemClient.clear_system_cache()
     return retriever