You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
swarms/swarms/structs/ui/ui.py

665 lines
26 KiB

import os
from dotenv import load_dotenv
from typing import List, Dict, Any, Tuple, Optional
import json
import time
import asyncio
import gradio as gr
from swarms.structs.agent import Agent
from swarms.structs.swarm_router import SwarmRouter
from swarms.utils.loguru_logger import initialize_logger
from swarm_models import OpenAIChat
# Initialize logger
logger = initialize_logger(log_folder="swarm_ui")
# Load environment variables
load_dotenv()
# Get the OpenAI API key from the environment variable
api_key = os.getenv("GROQ_API_KEY")
# changed to swarms_models
# adding functionality to view other models of swarms models
# Model initialization
model = OpenAIChat(
openai_api_base="https://api.groq.com/openai/v1",
openai_api_key=api_key,
model_name="llama-3.1-70b-versatile",
temperature=0.1,
)
# Define the path to agent_prompts.json
# locates the json file and then fetches the promopts
PROMPT_JSON_PATH = os.path.join(os.path.dirname(os.path.abspath(__file__)), "agent_prompts.json")
logger.info(f"Loading prompts from: {PROMPT_JSON_PATH}")
def load_prompts_from_json() -> Dict[str, str]:
try:
if not os.path.exists(PROMPT_JSON_PATH):
# Load default prompts
return {
"agent.data_extractor": "You are a data extraction agent...",
"agent.summarizer": "You are a summarization agent...",
"agent.onboarding_agent": "You are an onboarding agent..."
}
with open(PROMPT_JSON_PATH, 'r', encoding='utf-8') as f:
try:
data = json.load(f)
except json.JSONDecodeError:
# Load default prompts
return {
"agent.data_extractor": "You are a data extraction agent...",
"agent.summarizer": "You are a summarization agent...",
"agent.onboarding_agent": "You are an onboarding agent..."
}
if not isinstance(data, dict):
# Load default prompts
return {
"agent.data_extractor": "You are a data extraction agent...",
"agent.summarizer": "You are a summarization agent...",
"agent.onboarding_agent": "You are an onboarding agent..."
}
prompts = {}
for agent_name, details in data.items():
if not isinstance(details, dict) or "system_prompt" not in details:
continue
prompts[f"agent-{agent_name}"] = details["system_prompt"]
if not prompts:
# Load default prompts
return {
"agent.data_extractor": "You are a data extraction agent...",
"agent.summarizer": "You are a summarization agent...",
"agent.onboarding_agent": "You are an onboarding agent..."
}
return prompts
except Exception:
# Load default prompts
return {
"agent.data_extractor": "You are a data extraction agent...",
"agent.summarizer": "You are a summarization agent...",
"agent.onboarding_agent": "You are an onboarding agent..."
}
AGENT_PROMPTS = load_prompts_from_json()
def initialize_agents(dynamic_temp: float, agent_keys: List[str]) -> List[Agent]:
agents = []
seen_names = set()
for agent_key in agent_keys:
if agent_key not in AGENT_PROMPTS:
raise ValueError(f"Invalid agent key: {agent_key}")
agent_prompt = AGENT_PROMPTS[agent_key]
agent_name = agent_key.split('.')[-1]
# Ensure unique agent names
base_name = agent_name
counter = 1
while agent_name in seen_names:
agent_name = f"{base_name}_{counter}"
counter += 1
seen_names.add(agent_name)
agent = Agent(
agent_name=f"Agent-{agent_name}",
system_prompt=agent_prompt,
llm=model,
max_loops=1,
autosave=True,
verbose=True,
dynamic_temperature_enabled=True,
saved_state_path=f"agent_{agent_name}.json",
user_name="pe_firm",
retry_attempts=1,
context_length=200000,
output_type="string", # here is the output type which is string
temperature=dynamic_temp,
)
agents.append(agent)
return agents
async def execute_task(task: str, max_loops: int, dynamic_temp: float,
swarm_type: str, agent_keys: List[str], flow: str = None) -> Tuple[Any, 'SwarmRouter', str]:
"""
Enhanced task execution with comprehensive error handling and raw result return.
"""
try:
# Initialize agents
try:
agents = initialize_agents(dynamic_temp, agent_keys)
except Exception as e:
return None, None, str(e)
# Swarm-specific configurations
router_kwargs = {
"name": "multi-agent-workflow",
"description": f"Executing {swarm_type} workflow",
"max_loops": max_loops,
"agents": agents,
"autosave": True,
"return_json": True,
"output_type": "string",
"swarm_type": swarm_type, # Pass swarm_type here
}
if swarm_type == "AgentRearrange":
if not flow:
return None, None, "Flow configuration is required for AgentRearrange"
router_kwargs["flow"] = flow
if swarm_type == "MixtureOfAgents":
if len(agents) < 2:
return None, None, "MixtureOfAgents requires at least 2 agents"
if swarm_type == "SpreadSheetSwarm":
# spread sheet swarm needs specific setup
output_dir = "swarm_outputs"
os.makedirs(output_dir, exist_ok=True)
# Create a sanitized filename using only safe characters
timestamp = time.strftime("%Y%m%d_%H%M%S")
output_file = f"swarm_output_{timestamp}.csv"
output_path = os.path.join(output_dir, output_file)
# Validate the output path
try:
# Ensure the path is valid and writable
with open(output_path, 'w') as f:
pass
os.remove(output_path) # Clean up the test file
except OSError as e:
return None, None, str(e)
router_kwargs["output_path"] = output_path
# Create and execute SwarmRouter
try:
timeout = 450 if swarm_type != "SpreadSheetSwarm" else 900 # SpreadSheetSwarm will have different timeout.
router = SwarmRouter(**router_kwargs)
if swarm_type == "AgentRearrange":
result = await asyncio.wait_for(
asyncio.to_thread(router._run, task),
timeout=timeout
)
return result, router, ""
result = await asyncio.wait_for(
asyncio.to_thread(router.run, task),
timeout=timeout
)
if swarm_type == "SpreadSheetSwarm":
# Verify the output file was created
if not os.path.exists(output_path):
return None, None, "Output file was not created"
return output_path, router, ""
return result, router, ""
except asyncio.TimeoutError:
return None, None, f"Task execution timed out after {timeout} seconds"
except Exception as e:
return None, None, str(e)
except Exception as e:
return None, None, str(e)
class UI:
def __init__(self, theme):
self.theme = theme
self.blocks = gr.Blocks(theme=self.theme)
self.components = {} # Dictionary to store UI components
def create_markdown(self, text, is_header=False):
if is_header:
markdown = gr.Markdown(f"<h1 style='color: #ffffff; text-align: center;'>{text}</h1>")
else:
markdown = gr.Markdown(f"<p style='color: #cccccc; text-align: center;'>{text}</p>")
self.components[f'markdown_{text}'] = markdown
return markdown
def create_text_input(self, label, lines=3, placeholder=""):
text_input = gr.Textbox(
label=label,
lines=lines,
placeholder=placeholder,
elem_classes=["custom-input"],
)
self.components[f'text_input_{label}'] = text_input
return text_input
def create_slider(self, label, minimum=0, maximum=1, value=0.5, step=0.1):
slider = gr.Slider(
minimum=minimum,
maximum=maximum,
value=value,
step=step,
label=label,
interactive=True,
)
self.components[f'slider_{label}']
return slider
def create_dropdown(self, label, choices, value=None, multiselect=False):
if not choices:
choices = ["No options available"]
if value is None and choices:
value = choices[0] if not multiselect else [choices[0]]
dropdown = gr.Dropdown(
label=label,
choices=choices,
value=value,
interactive=True,
multiselect=multiselect,
)
self.components[f'dropdown_{label}'] = dropdown
return dropdown
def create_button(self, text, variant="primary"):
button = gr.Button(text, variant=variant)
self.components[f'button_{text}'] = button
return button
def create_text_output(self, label, lines=10, placeholder=""):
text_output = gr.Textbox(
label=label,
interactive=False,
placeholder=placeholder,
lines=lines,
elem_classes=["custom-output"],
)
self.components[f'text_output_{label}'] = text_output
return text_output
def create_tab(self, label, content_function):
with gr.Tab(label):
content_function(self)
def set_event_listener(self, button, function, inputs, outputs):
button.click(function, inputs=inputs, outputs=outputs)
def get_components(self, *keys):
if not keys:
return self.components # return all components
return [self.components[key] for key in keys]
def create_json_output(self, label, placeholder=""):
json_output = gr.JSON(
label=label,
value={},
elem_classes=["custom-output"],
)
self.components[f'json_output_{label}'] = json_output
return json_output
def build(self):
return self.blocks
def create_conditional_input(self, component, visible_when, watch_component):
"""Create an input that's only visible under certain conditions"""
watch_component.change(
fn=lambda x: gr.update(visible=visible_when(x)),
inputs=[watch_component],
outputs=[component]
)
@staticmethod
def create_ui_theme(primary_color="red"):
return gr.themes.Soft(
primary_hue=primary_color,
secondary_hue="gray",
neutral_hue="gray",
).set(
body_background_fill="#20252c",
body_text_color="#f0f0f0",
button_primary_background_fill=primary_color,
button_primary_text_color="#ffffff",
button_secondary_background_fill=primary_color,
button_secondary_text_color="#ffffff",
shadow_drop="0px 2px 4px rgba(0, 0, 0, 0.3)",
)
def create_agent_details_tab(self):
"""Create the agent details tab content."""
with gr.Column():
gr.Markdown("### Agent Details")
gr.Markdown("""
**Available Agent Types:**
- Data Extraction Agent: Specialized in extracting relevant information
- Summary Agent: Creates concise summaries of information
- Analysis Agent: Performs detailed analysis of data
**Swarm Types:**
- ConcurrentWorkflow: Agents work in parallel
- SequentialWorkflow: Agents work in sequence
- AgentRearrange: Custom agent execution flow
- MixtureOfAgents: Combines multiple agents with an aggregator
- SpreadSheetSwarm: Specialized for spreadsheet operations
- Auto: Automatically determines optimal workflow
""")
return gr.Column()
def create_logs_tab(self):
"""Create the logs tab content."""
with gr.Column():
gr.Markdown("### Execution Logs")
logs_display = gr.Textbox(
label="System Logs",
placeholder="Execution logs will appear here...",
interactive=False,
lines=10
)
return logs_display
def create_app():
# Initialize UI
theme = UI.create_ui_theme(primary_color="red")
ui = UI(theme=theme)
with ui.blocks:
with gr.Row():
with gr.Column(scale=4): # Left column (80% width)
ui.create_markdown("Swarms", is_header=True)
ui.create_markdown(
"<b>The Enterprise-Grade Production-Ready Multi-Agent Orchestration Framework</b>"
)
with gr.Row():
with gr.Column(scale=4):
with gr.Row():
task_input = gr.Textbox(
label="Task Description",
placeholder="Describe your task here...",
lines=3
)
with gr.Row():
with gr.Column(scale=1):
# Get available agent prompts
available_prompts = list(AGENT_PROMPTS.keys()) if AGENT_PROMPTS else ["No agents available"]
agent_prompt_selector = gr.Dropdown(
label="Select Agent Prompts",
choices=available_prompts,
value=[available_prompts[0]] if available_prompts else None,
multiselect=True,
interactive=True
)
with gr.Column(scale=1):
# Get available swarm types
swarm_types = [
"SequentialWorkflow", "ConcurrentWorkflow", "AgentRearrange",
"MixtureOfAgents", "SpreadSheetSwarm", "auto"
]
agent_selector = gr.Dropdown(
label="Select Swarm",
choices=swarm_types,
value=swarm_types[0],
multiselect=False,
interactive=True
)
# Flow configuration components for AgentRearrange
with gr.Column(visible=False) as flow_config:
flow_text = gr.Textbox(
label="Agent Flow Configuration",
placeholder="Enter agent flow (e.g., Agent1 -> Agent2 -> Agent3)",
lines=2
)
gr.Markdown(
"""
**Flow Configuration Help:**
- Enter agent names separated by ' -> '
- Example: Agent1 -> Agent2 -> Agent3
- Use exact agent names from the prompts above
"""
)
with gr.Column(scale=2, min_width=200):
with gr.Row():
max_loops_slider = gr.Slider(
label="Max Loops",
minimum=1,
maximum=10,
value=1,
step=1
)
with gr.Row():
dynamic_slider = gr.Slider(
label="Dynamic Temp",
minimum=0,
maximum=1,
value=0.1,
step=0.01
)
with gr.Row():
loading_status = gr.Textbox(
label="Status",
value="Ready",
interactive=False
)
with gr.Row():
run_button = gr.Button("Run Task", variant="primary")
cancel_button = gr.Button("Cancel", variant="secondary")
# Add loading indicator and status
with gr.Row():
agent_output_display = gr.Textbox(
label="Agent Responses",
placeholder="Responses will appear here...",
interactive=False,
lines=10
)
def update_flow_agents(agent_keys):
"""Update flow agents based on selected agent prompts."""
if not agent_keys:
return [], "No agents selected"
agent_names = [key.split('.')[-1] for key in agent_keys]
return agent_names, "Select agents in execution order"
def update_flow_preview(selected_flow_agents):
"""Update flow preview based on selected agents."""
if not selected_flow_agents:
return "Flow will be shown here..."
flow = " -> ".join(selected_flow_agents)
return flow
def update_ui_for_swarm_type(swarm_type):
"""Update UI components based on selected swarm type."""
is_agent_rearrange = swarm_type == "AgentRearrange"
is_mixture = swarm_type == "MixtureOfAgents"
is_spreadsheet = swarm_type == "SpreadSheetSwarm"
max_loops = 5 if is_mixture or is_spreadsheet else 10
# Return visibility state for flow configuration and max loops update
return (
gr.update(visible=is_agent_rearrange), # For flow_config
gr.update(maximum=max_loops), # For max_loops_slider
f"Selected {swarm_type}" # For loading_status
)
async def run_task_wrapper(task, max_loops, dynamic_temp, swarm_type, agent_prompt_selector, flow_text):
"""
Execute the task and update the UI with progress, saving the raw AgentRearrange response
and parsing it for display.
"""
try:
if not task:
yield "Please provide a task description.", "Error: Missing task"
return
if not agent_prompt_selector or len(agent_prompt_selector) == 0:
yield "Please select at least one agent.", "Error: No agents selected"
return
# Update status
yield "Processing...", "Running task..."
# Prepare flow for AgentRearrange
flow = None
if swarm_type == "AgentRearrange":
if not flow_text:
yield "Please provide the agent flow configuration.", "Error: Flow not configured"
return
flow = flow_text
# Execute the task
result, router, error = await execute_task(
task=task,
max_loops=max_loops,
dynamic_temp=dynamic_temp,
swarm_type=swarm_type,
agent_keys=agent_prompt_selector,
flow=flow
)
if error:
yield f"Error: {error}", "Error occurred"
return
# Process result based on swarm type
if swarm_type == "AgentRearrange":
# Store raw response in a temporary JSON file
temp_json_path = "temp_agent_rearrange.json"
with open(temp_json_path, "w", encoding="utf-8") as temp_file:
json.dump(result, temp_file, indent=4)
# Read from the temporary JSON file
with open(temp_json_path, "r", encoding="utf-8") as temp_file:
temp_json = json.load(temp_file)
# Parse and format the JSON output
formatted_output = parse_agent_rearrange_output(temp_json)
yield formatted_output, "Completed"
return
# Generic processing for other swarm types
output_lines = []
if isinstance(result, dict): # For JSON-like outputs
for key, value in result.items():
output_lines.append(f"### {key} ###\n{value}\n{'=' * 50}\n")
elif isinstance(result, str):
output_lines.append(result)
yield "\n".join(output_lines), "Completed"
except Exception as e:
yield f"Error: {str(e)}", "Error occurred"
def parse_agent_rearrange_output(raw_json: dict) -> str:
"""
Parse the AgentRearrange JSON output and format it for display.
"""
output_lines = []
# Input Section
input_data = raw_json.get("input", {})
swarm_id = input_data.get("swarm_id", "N/A")
swarm_name = input_data.get("name", "N/A")
flow = input_data.get("flow", "N/A")
output_lines.append(f"### Swarm ID: {swarm_id} ###")
output_lines.append(f"### Swarm Name: {swarm_name} ###")
output_lines.append(f"### Flow: {flow} ###")
# Outputs Section
outputs = raw_json.get("outputs", [])
for i, agent_output in enumerate(outputs, start=1):
agent_name = agent_output.get("agent_name", "N/A")
task = agent_output.get("task", "N/A")
output_lines.append(f"\n#### Agent: {agent_name} (Step {i}) ####")
output_lines.append(f"Task: {task}")
# Steps Section
steps = agent_output.get("steps", [])
for step_idx, step in enumerate(steps, start=1):
role = step.get("role", "N/A")
content = step.get("content", "N/A")
output_lines.append(f" - **Loop 1: Role:** {role} **Content:** {content}")
output_lines.append(f"{'=' * 30}\n")
return "\n".join(output_lines)
# Connect the update functions
agent_selector.change(
fn=update_ui_for_swarm_type,
inputs=[agent_selector],
outputs=[flow_config, max_loops_slider, loading_status]
)
# Create event trigger
# Create event trigger for run button
run_event = run_button.click(
fn=run_task_wrapper,
inputs=[
task_input,
max_loops_slider,
dynamic_slider,
agent_selector,
agent_prompt_selector,
flow_text
],
outputs=[agent_output_display, loading_status]
)
# Connect cancel button to interrupt processing
def cancel_task():
return "Task cancelled.", "Cancelled"
cancel_button.click(
fn=cancel_task,
inputs=None,
outputs=[agent_output_display, loading_status],
cancels=run_event
)
with gr.Column(scale=1): # Right column
with gr.Tabs():
with gr.Tab("Agent Details"):
ui.create_agent_details_tab()
with gr.Tab("Logs"):
logs_display = ui.create_logs_tab()
def update_logs_display():
"""Update logs display with current logs."""
return ""
# Update logs when tab is selected
logs_tab = gr.Tab("Logs")
logs_tab.select(fn=update_logs_display, inputs=None, outputs=[logs_display])
return ui.build()
if __name__ == "__main__":
app = create_app()
app.launch()