aya-chat-app.py

from crewai import Agent, Task, Crew, Process, LLM
import asyncio
from textwrap import dedent
import os
import litellm
import uuid
litellm.set_verbose=False

from pathlib import Path
log_file = "/tmp/crew.log"
Path(log_file).touch()

llm = LLM(model="bedrock/" + os.getenv("AWS_BEDROCK_MODEL"))
print("Importing available tools")
import tools.tool_ticket_query as tool_ticket_query
ticket_search_tool = tool_ticket_query.TicketListingTool()

import tools.tool_customer_consumption as tool_customer_consumption
consumption_change_tool = tool_customer_consumption.ConsumptionMetricsTool()

import tools.tool_accounts_kg as knowledge_graph_accounts
account_info_tool = knowledge_graph_accounts.ToolAccountsKnowledgeGraph()

import tools.tool_case_summarizer as tool_case_summarizer
case_summarizer_tool = tool_case_summarizer.ToolCaseSummarizer()

# Agent Definitions
agent_1 = Agent(
    role=dedent((
        """
        Customer information lookup assistant.
        """)), # Think of this as the job title
    backstory=dedent((
        """
        You are an advanced Cloudera Customer Enablement assistant, attempting to proivde information about CML Customers, their deployed workspaces and entitlements, and usage information within those workspaces.
        This may include lookups of filed support cases, summaries of support cases, overall account consumption, workload consumption trends, and field information.
        """)), # This is the backstory of the agent, this helps the agent to understand the context of the task
    goal=dedent((
        """
        Perform the lookup task assigned to you by appropriate lookup tools and explain. Try to keep final answers in markdown format.
        """)), # This is the goal that the agent is trying to achieve
    tools=[ticket_search_tool,
           case_summarizer_tool,
           consumption_change_tool,
           account_info_tool],
    allow_delegation=False,
    max_iter=1,
    max_retry_limit=3,
    max_retries=3,
    llm=llm,
    verbose=True
)

# Task Definitions
import datetime
print("Defining Primary Task")

task_1 = Task(
    description=dedent((
        """
        Attempt to answer the user question about the customer.
        ---
        Request ID: "{req_id}"
        User Input: "{req_input}"
        """)),
    expected_output=dedent((
        """
        Output should be a well formatted list or statement with the results of the user request.
        """)),
    agent=agent_1,
)

def crew_launch(req_id, req_input):
    # Instantiate your crew with a sequential process
    print("Instantiating Crew")
    crew = Crew(
        agents=[agent_1],
        tasks=[task_1],
        verbose=True,  # You can set it to True or False
        # ↑ indicates the verbosity level for logging during execution.
        process=Process.sequential
    )
    print("Setting req_input")
    inputs = {
        "req_id": req_id,
        "req_input": req_input,
    }
    print("Kicking off crew")
    result = crew.kickoff(inputs=inputs)
    print(result.tasks_output)
    return result
  
def read_logs():
    with open(log_file, "r") as f:
        return f.read()


example=""
print("Loading UI")
import gradio as gr


bot_msg = """<strong style="text-align:left;">A.Y.A.</strong> - %s\n\n%s"""
#human_msg="##### User\n%s"
human_msg="""<p align="right">%s - <strong>User</strong></p>\n\n%s"""

startup_history = [(None, bot_msg % (datetime.datetime.now().strftime('%H:%M'), "Hello, how can I help you today?"))]




def display_user_message(message, chat_history):
    request_id = str(uuid.uuid4())[:8]
    message_text = message["text"]
    chat_history.append((human_msg % (datetime.datetime.now().strftime('%H:%M'), message_text), None))
    return request_id, message_text, chat_history

def display_thinking(chat_history):
    thinking_msg = """
<h3 style="text-align:left;">🔄 Thinking...</h3>

"""

    chat_history.append((None, thinking_msg))
    return chat_history

def respond(request_id, message_text, chat_history):
    agent_usage_template = """
<h3 style="text-align:left;">🛠️ Calling Agent ...</h3>

`%s`
"""
    crew_response = crew_launch(request_id, message_text)
    try:
        tool_usage_file = open("/tmp/%s" % request_id , "r")
        tool_usage_text = tool_usage_file.read()
    except:
        tool_usage_text = "Unknown Agent"
    agent_usage_message = agent_usage_template % tool_usage_text
    chat_history.append((None, agent_usage_message))
    chat_history.append((None, bot_msg % (datetime.datetime.now().strftime('%H:%M'), str(crew_response))))
    return chat_history

def maybe_update_status(status, chat_history):
    crew_status = status
    chat_history.append((None, bot_msg % crew_status))

css = """
footer{display:none !important}
#examples_table {zoom: 70% !important;}
#chatbot { flex-grow: 1 !important; overflow: auto !important;}
#col { height: 75vh !important; }
.info_md .container {
    border:1px solid #ccc; 
    border-radius:5px; 
    min-height:300px;
    color: #666;
    padding: 10px;
    background-color: whitesmoke;
}
"""

header_text = """
# All Your Agents
Meet **A.Y.A**, an Agentic Workflow Orchestrator which deciphers and sends User requests to domain specific AI Agents and Tools.
"""

info_text = """
<div class='container'> 

## Agents A.Y.A. Can Use
**📝 Ticket Database Agent**
                               
Queries a db for recent customer incident tickets.

##### 📖 Support Case Summarization Agent
                               
AI Summarization Agent for support case details and comments.

##### 📈 Consumption Metrics Agent
                               
Looks up recent trends in Customer Consumption.

##### 👤 Account Personnel Registry Agent
                               
Knowledge graph lookup of Customer Account Team personnel.
</div>
"""

theme = gr.themes.Base().set(
    body_background_fill="url('file=/home/cdsw/assets/background.png') #FFFFFF no-repeat center bottom / 100svw auto padding-box fixed"
)

# Define this textbox outside of blocks so other components can refer to it, render it on the layout inside gr.Blocks
input = gr.MultimodalTextbox(scale = 5, show_label = False, file_types = ["text"])

with gr.Blocks(css=css, theme=theme, title="A.Y.A.") as demo:
    request_id = gr.State("")
    request_text = gr.State("")
    with gr.Row():
        gr.Markdown(header_text)
    with gr.Row():
        with gr.Column(scale=6):
            info = gr.Markdown(info_text, elem_classes=["info_md"])
            example_num = gr.Textbox(visible = False)
            with gr.Accordion("Example User Inputs", open = True):
              # TODO: Re-add kg agent example when ready
              #[4, {"text":"Who are the key personnel managing the customer Evolve Pharma"}],
              examples_2 = gr.Examples([
                                          [1, {"text":"Show me the weekly average consumption changes for customers"}],
                                          [2, {"text":"Show me recent incident tickets from Evolve Pharma"}],
                                          [3, {"text":"Summarize support case ID 981492 in markdown"}],
                                          [4, {"text":"Who are the support personnel responsible for Evolve Pharma?"}]
                                      ],
                                      inputs=[example_num, input], elem_id="examples_table", label="")
        with gr.Column(scale=15, elem_id="col"):
            chatbot = gr.Chatbot(
                value = startup_history,
                avatar_images=["assets/person.png", "assets/chatbot.png"],
                elem_id = "chatbot",
                show_label = False
            )
            input.render()
            #clear = gr.ClearButton(scale = 1, components = [chatbot], value = "Clear History")
    user_msg = input.submit(display_user_message, [input, chatbot],  [request_id, request_text, chatbot])
    input.submit(lambda x: gr.update(value={"text":""}), None, [input], queue=False)
    tool_msg = user_msg.then(display_thinking, chatbot, chatbot)
    ayb_msg = tool_msg.then(respond, [request_id, request_text, chatbot], chatbot)

demo.launch(server_port=int(os.getenv("CDSW_APP_PORT")), server_name="127.0.0.1",  debug=True, allowed_paths=["/home/cdsw/assets/background.png"])