mealgraph / mealgraph.py
moazeldegwy's picture
Simplify topology to 3 agents + 2 tools
1933348
import json
import os
from datetime import datetime
from typing import Any, Dict, List, Optional
from IPython.display import Markdown, display
from agents import CoachAgent, MedicalAssessmentAgent, PlannerAgent
from config import set_settings
from logging_setup import get_logger, refresh_level
from memory import LongTermMemory
from state import initialize_empty_memory
from tools import QuantitiesFinder, WebSearchTool
from utils import APIPoolManager, create_llm
from workflow import setup_workflow as setup_workflow_workflow
_logger = get_logger("mealgraph")
def debug(level: str = "full", scopes: Optional[Dict[str, List[str]]] = None) -> None:
"""Enable debug mode with the given level and scopes.
Args:
level: 'full' (default) to show inputs and outputs, or 'output' to show only outputs.
scopes: Optional dict like ``{'agents': ['all'], 'tools': ['QuantitiesFinder']}``.
If None, defaults to all agents and tools.
"""
if scopes is None:
scopes = {"agents": ["all"], "tools": ["all"]}
set_settings(debug_mode=True, debug_level=level, debug_scopes=scopes)
refresh_level()
def logging(log_dir: Optional[str] = None, persistence_dir: Optional[str] = None) -> None: # noqa: A001 - public name kept for backwards compat
"""Set directories for log files and LangGraph checkpoint persistence.
If ``log_dir`` is provided, agent/tool I/O is dumped there as JSON.
If ``persistence_dir`` is provided, LangGraph checkpoints are persisted to disk.
If neither is set, logging is disabled and persistence is in-memory.
"""
updates: Dict[str, Any] = {}
if log_dir is not None:
os.makedirs(log_dir, exist_ok=True)
updates["log_dir"] = log_dir
if persistence_dir is not None:
os.makedirs(persistence_dir, exist_ok=True)
updates["persistence_dir"] = persistence_dir
if updates:
set_settings(**updates)
# Default model configurations (without API keys, as they will be provided
# by the user). Five LLM slots — no separate validation_agent any more
# since the Validator was folded into the Planner / Coach.
#
# Targets the Gemini 3.x family via the rolling "*-latest" aliases:
# - gemini-pro-latest (deep reasoning: Coach / Medical / Planner)
# - gemini-flash-latest (mid-tier; reserved for overrides)
# - gemini-flash-lite-latest (cheapest; tools + simulator)
DEFAULT_MODEL_CONFIGS = {
"main": {
"type": "gemini",
"model_name": "gemini-pro-latest",
"structured_output": True,
"thinking_budget": 600,
"params": {"max_tokens": 5120, "temperature": 0.3},
},
"agents_llm": {
"type": "gemini",
"model_name": "gemini-pro-latest",
"structured_output": True,
"thinking_budget": 600,
"params": {"max_tokens": 5120, "temperature": 0.3},
},
"planner_agent": {
"type": "gemini",
"model_name": "gemini-pro-latest",
"structured_output": True,
"thinking_budget": 600,
"params": {"max_tokens": 5120, "temperature": 0.3},
},
"tools_llm": {
"type": "gemini",
"model_name": "gemini-flash-lite-latest",
"structured_output": False,
"thinking_budget": 600,
"params": {"max_tokens": 5120, "temperature": 0.3},
},
"user_simulator": {
"type": "gemini",
"model_name": "gemini-flash-lite-latest",
"structured_output": False,
"thinking_budget": 300,
"params": {"max_tokens": 5120, "temperature": 0.5},
},
}
# Global variables to hold the system components
LLM_INSTANCES = None
TOOLS = None
AGENTS = None
APP = None
def create_llm_instances(
api_keys: list[str],
model_overrides: Optional[Dict[str, Any]] = None,
enable_rate_limiting: bool = True,
):
"""Create LLM instances using provided API keys list and optional model overrides.
Args:
api_keys: List of API keys to cycle through.
model_overrides: Optional overrides for model configs.
enable_rate_limiting: If True, apply rate limits (default). If False, disable rate limiting and just cycle keys.
"""
global LLM_INSTANCES
if not api_keys:
raise ValueError("At least one API key must be provided.")
if enable_rate_limiting:
rate_limits = {
"gemini-pro-latest": (5, 100),
"gemini-flash-latest": (10, 250),
"gemini-flash-lite-latest": (15, 500),
}
else:
rate_limits = None
manager = APIPoolManager(api_keys, rate_limits)
_logger.info(
"APIPoolManager initialized with %s and %d API keys.",
"rate limiting enabled" if enable_rate_limiting else "rate limiting disabled",
len(api_keys),
)
# ``cfg`` (not ``config``) — the latter would shadow the imported
# :mod:`config` module inside the loop body.
model_configs: Dict[str, Dict[str, Any]] = {}
for key in DEFAULT_MODEL_CONFIGS:
cfg = DEFAULT_MODEL_CONFIGS[key].copy()
if model_overrides and key in model_overrides:
override = model_overrides[key]
if "model_name" in override:
cfg["model_name"] = override["model_name"]
if "params" in override:
cfg["params"] = {**cfg.get("params", {}), **override["params"]}
model_configs[key] = cfg
LLM_INSTANCES = {
"main": create_llm(model_configs["main"], manager),
"agents_llm": create_llm(model_configs["agents_llm"], manager),
"planner_agent": create_llm(model_configs["planner_agent"], manager),
"tools_llm": create_llm(model_configs["tools_llm"], manager),
"user_simulator": create_llm(model_configs["user_simulator"], manager),
}
def initialize_tools():
"""Initialize tools using the LLM instances."""
global TOOLS
if LLM_INSTANCES is None:
raise RuntimeError("LLM instances must be created before initializing tools.")
TOOLS_LLM = LLM_INSTANCES["tools_llm"]
TOOLS = {
"WebSearchTool": WebSearchTool(TOOLS_LLM),
"QuantitiesFinder": QuantitiesFinder(),
}
def initialize_agents():
"""Initialize agents using the LLM instances and tools."""
global AGENTS
if LLM_INSTANCES is None or TOOLS is None:
raise RuntimeError("LLM instances and tools must be initialized before agents.")
MAIN_LLM = LLM_INSTANCES["main"]
AGENTS_LLM = LLM_INSTANCES["agents_llm"]
PLANNER_LLM = LLM_INSTANCES["planner_agent"]
AGENTS = {
"CoachAgent": CoachAgent(MAIN_LLM),
"MedicalAssessmentAgent": MedicalAssessmentAgent(
AGENTS_LLM, TOOLS["WebSearchTool"]
),
"PlannerAgent": PlannerAgent(
PLANNER_LLM, TOOLS["WebSearchTool"], TOOLS["QuantitiesFinder"]
),
}
# ---------------------------------------------------------------------------
# Long-term memory singleton
# ---------------------------------------------------------------------------
LONG_TERM_MEMORY: Optional[LongTermMemory] = None
def initialize_long_term_memory(db_path: Optional[str] = None) -> LongTermMemory:
"""Initialise the SQLite-backed three-tier memory.
Pass a file path for cross-session persistence, or omit for an in-memory
DB (default; tests / ephemeral demos).
"""
global LONG_TERM_MEMORY
LONG_TERM_MEMORY = LongTermMemory(db_path=db_path)
_logger.info("Long-term memory initialised at %s", db_path or ":memory:")
return LONG_TERM_MEMORY
def setup_workflow():
global APP
if AGENTS is None or TOOLS is None:
raise RuntimeError("Agents and tools must be initialized before setting up workflow.")
APP = setup_workflow_workflow(AGENTS["CoachAgent"], AGENTS, TOOLS)
class UserSimulator:
def __init__(self, llm, user_profile, medical_history):
self.llm = llm
self.user_data = {
"user_profile": user_profile,
"medical_history": medical_history,
}
def get_response(self, assistant_message):
user_data_str = json.dumps(self.user_data, indent=2)
sim_prompt = f"""You are a user interacting with a nutrition app. Here is your profile and medical history:
{user_data_str}
The app has asked: {assistant_message}
Provide a realistic answer based on your profile and medical history.
Your response:
"""
response = self.llm(sim_prompt)[0]
return response
def initialize_user_data():
"""Collect user information interactively to initialize memory."""
print("Let's collect some information about you to personalize your experience.")
print("\nFirst, your demographics and anthropometrics:")
name = input("What is your name? ")
age = float(input("How old are you? (in years) "))
sex = input("What is your sex? (male/female) ")
height = float(input("What is your height? (in cm) "))
weight = float(input("What is your weight? (in kg) "))
print("\nNext, about your lifestyle and goals:")
activity_level = input(
"What is your activity level? (e.g., sedentary, lightly active, "
"moderately active, very active, extra active) "
)
goal = input(
"What is your primary goal? (e.g., lose weight, maintain weight, gain muscle) "
)
job = input("What is your job or daily routine? ")
print("\nDietary preferences:")
dietary_restrictions = input(
"Any dietary restrictions? (e.g., vegetarian, vegan, keto) "
)
food_likes = input("Favorite foods? ")
food_dislikes = input("Foods you dislike? ")
allergies_input = input("Any allergies? (comma-separated) ")
allergies_list = (
[a.strip() for a in allergies_input.split(",") if a.strip()]
if allergies_input
else []
)
print("\nLocation and budget:")
country = input("Which country are you in? ")
currency = input("Preferred currency? (e.g., USD, EGP) ")
print("\nMedical history:")
conditions_input = input("Any medical conditions? (comma-separated) ")
conditions_list = (
[c.strip() for c in conditions_input.split(",") if c.strip()]
if conditions_input
else []
)
medications_input = input("Current medications? (comma-separated) ")
medications_list = (
[m.strip() for m in medications_input.split(",") if m.strip()]
if medications_input
else []
)
past_issues_input = input("Past health issues? (comma-separated) ")
past_issues_list = (
[p.strip() for p in past_issues_input.split(",") if p.strip()]
if past_issues_input
else []
)
lab_results = input("Any recent lab results? (e.g., cholesterol levels) ")
user_profile = {
"name": name,
"age": age,
"sex": sex,
"height": height,
"weight": weight,
"activity_level": activity_level,
"goal": goal,
"job": job,
"dietary_restrictions": dietary_restrictions,
"food_likes": food_likes,
"food_dislikes": food_dislikes,
"allergies": allergies_list,
"country": country,
"currency": currency,
"last_updated": datetime.now().isoformat(),
}
medical_history = {
"conditions": conditions_list,
"medications": medications_list,
"past_issues": past_issues_list,
"lab_results": lab_results,
"last_updated": datetime.now().isoformat(),
}
memory = initialize_empty_memory()
memory["user_profile"] = user_profile
memory["medical_history"] = medical_history
return memory
def run(simulate=False, simulated_users=None):
"""Run the system in either simulation or interactive mode."""
if APP is None:
raise RuntimeError("Workflow must be set up before running the system.")
if simulate:
print("\n" + "=" * 80)
print("STARTING SIMULATION MODE")
print("=" * 80)
if not simulated_users:
raise ValueError("simulated_users must be provided when simulate=True")
for user in simulated_users:
print(f"\nProcessing user: {user['user_profile']['name']}")
user["user_profile"]["last_updated"] = datetime.now().isoformat()
user["medical_history"]["last_updated"] = datetime.now().isoformat()
memory = {
"user_profile": user["user_profile"],
"medical_history": user["medical_history"],
"flags_and_assessments": {},
"plans": {},
}
conversation_history = []
previous_actions = []
user_simulator = UserSimulator(
LLM_INSTANCES["user_simulator"],
user["user_profile"],
user["medical_history"],
)
for question in user["questions"]:
print(f"\n🙍🏻‍♂️Asking: {question}")
state = {
"memory": memory,
"user_question": question,
"conversation_history": conversation_history
+ [{"role": "user", "content": question}],
"current_action": None,
"agent_result": None,
"num_turns": 0,
"max_turns": 10,
"previous_actions": previous_actions,
"response_steps": [],
}
while True:
final_state = APP.invoke(
state,
config={
"configurable": {
"thread_id": f"user_{user['user_profile']['name']}"
}
},
)
if final_state["num_turns"] >= final_state["max_turns"]:
print("Max turns reached without composing a response.")
break
last_action = final_state["current_action"]["action"]
if last_action == "compose_response":
print(f"\n{'='*60}")
display(Markdown(f"**Answer:**\n\n{final_state['agent_result']}"))
conversation_history = final_state["conversation_history"]
memory = final_state["memory"]
previous_actions = final_state["previous_actions"]
break
elif last_action == "ask_user":
prompt = final_state["agent_result"]
print(f"System asks: {prompt}")
response = user_simulator.get_response(prompt)
print(f"Simulated user responds: {response}")
state = {
"memory": memory,
"user_question": question,
"conversation_history": conversation_history
+ [{"role": "user", "content": question}],
"current_action": None,
"agent_result": None,
"num_turns": 0,
"max_turns": 10,
"previous_actions": previous_actions,
"response_steps": [],
}
else:
print(f"Unexpected action: {last_action}")
break
else:
print("\n" + "=" * 80)
print("STARTING INTERACTIVE MODE")
print("=" * 80)
memory = initialize_user_data()
print("\n" + "=" * 80)
print("WELCOME TO THE NUTRITION APP")
print("=" * 80)
initial_state = {
"memory": memory,
"user_question": "welcome",
"conversation_history": [],
"current_action": None,
"agent_result": None,
"num_turns": 0,
"max_turns": 10,
"previous_actions": [],
"response_steps": [],
}
final_state = APP.invoke(initial_state, config={"configurable": {"thread_id": "user1"}})
if final_state["agent_result"]:
display(Markdown(f"\n🤖 Coach: {final_state['agent_result']}"))
memory = final_state["memory"]
conversation_history = final_state["conversation_history"]
while True:
q = input("\n❓ Your question: ")
if q.lower() == "exit":
break
state = {
"memory": final_state["memory"],
"user_question": q,
"conversation_history": final_state["conversation_history"]
+ [{"role": "user", "content": q}],
"current_action": None,
"agent_result": None,
"num_turns": 0,
"max_turns": 10,
"previous_actions": final_state["previous_actions"],
"response_steps": [],
}
final_state = APP.invoke(state, config={"configurable": {"thread_id": "user1"}})
if final_state["agent_result"]:
print(f"\n{'='*60}")
display(Markdown(f"\n🤖 Coach: {final_state['agent_result']}"))
memory = final_state["memory"]
conversation_history = final_state["conversation_history"]