Remo Multi-Agent Orchestration System

What is Remo?

Remo is a smart assistant that can help you with reminders, todos, emails, and creating images. It works like a team: there's a supervisor (Remo) who listens to you and then asks the right specialist (agent) to help. Each agent has special tools to do their job.

---

How Does Remo Work? (Step by Step)

1. You (the User) send a message (like "Remind me to call mom at 5pm").
2. Remo (the Supervisor) gets your message and figures out what you want.
3. Remo picks the right agent (like the Reminder Agent for reminders).
4. The agent uses its tools (like a function to save a reminder) to do the job.
5. The agent sends the answer back to Remo.
6. Remo wraps the answer in a friendly message and sends it back to you.

---

Main Technologies Used

1. LangChain

• A toolkit for building AI assistants that can use tools and remember things.
• Used for: Tool wrapping, memory, and agent logic (see src/agents/ and src/memory/).

2. LangGraph

• Lets you build workflows where different agents and tools can talk to each other in a graph (like a flowchart).
• Used for: Creating the supervisor and agent logic (see src/orchestration/ and agent files).

3. LangSmith

• (Not directly used, but supported by LangGraph/LangChain for tracing and analytics.)
• You can add it for debugging and tracking conversations.

4. AWS Bedrock

• Amazon's service for running powerful AI models (like chatbots and image generators).
• Used for: Running the main AI models that power Remo and the agents (see src/agents/content_creator/content_creator_tools.py and agent files).

5. DynamoDB

• Amazon's NoSQL database for saving data.
• Used for: Storing reminders, todos, emails, and conversation history (see src/utils/dynamodb_service.py).

6. Google Client

• Used for: Integrating with Google services (like Google Calendar for scheduling meetings). (See src/agents/email/email_tools.py and src/utils/google_calendar_service.py)

---

Architecture Diagram

flowchart TD
    User["User"] -->|"User prompt"| Supervisor["SupervisorOrchestrator (Remo)"]
    Supervisor -->|"Intent detection & routing"| Agents["Specialized Agents"]
    Agents --> ReminderAgent["ReminderAgent\n(create_react_agent)"]
    Agents --> TodoAgent["TodoAgent\n(create_react_agent)"]
    Agents --> EmailAgent["EmailAgent\n(create_react_agent)"]
    Agents --> ContentCreatorAgent["ContentCreatorAgent"]
    ReminderAgent --> ReminderTools["reminder_tools.py"]
    TodoAgent --> TodoTools["todo_tools.py"]
    EmailAgent --> EmailTools["email_tools.py"]
    ContentCreatorAgent --> ContentCreatorTools["content_creator_tools.py"]
    Supervisor -->|"Aggregates & wraps responses"| User
    subgraph LangGraph
        Supervisor
        ReminderAgent
        TodoAgent
        EmailAgent
    end
    classDef agent fill:#f9f,stroke:#333,stroke-width:2px;
    class ReminderAgent,TodoAgent,EmailAgent,ContentCreatorAgent agent;

Loading

---

Who Does What? (Agents and Their Tools)

1. SupervisorOrchestrator (Remo)

• Listens to the user.
• Decides which agent should help.
• Makes sure the answer is friendly and clear.
• Uses LangGraph to manage the flow.
• File: src/orchestration/supervisor.py

2. ReminderAgent

• Handles reminders (set, list, update, delete, mark complete).
• Tools: set_reminder, list_reminders, update_reminder, delete_reminder, mark_reminder_complete (in reminder_tools.py).
• Uses LangGraph's create_react_agent for smart reasoning.
• File: src/agents/reminders/reminder_agent.py

3. TodoAgent

• Handles todos (add, list, update, delete, mark complete, prioritize).
• Tools: add_todo, list_todos, update_todo, delete_todo, mark_todo_complete, prioritize_todos (in todo_tools.py).
• Uses LangGraph's create_react_agent.
• File: src/agents/todo/todo_agent.py

4. EmailAgent

• Handles emails (compose, send, schedule, search, mark read, archive, forward, reply, summarize).
• Tools: compose_email, send_email, schedule_email, search_emails, mark_email_read, archive_email, forward_email, reply_to_email, get_email_summary (in email_tools.py).
• Can also schedule meetings using Google Calendar.
• Uses LangGraph's create_react_agent.
• File: src/agents/email/email_agent.py

5. ContentCreatorAgent

• Handles image generation (and in the future, video).
• Tool: generate_nova_canvas_image (in content_creator_tools.py).
• File: src/agents/content_creator/content_creator_agent.py

---

DynamoDB Tables (and What's Inside)

• remo-reminders: Stores reminders for each user.
• remo-todos: Stores todos for each user.
• remo-conversations: Stores conversation history (each message with user_id, timestamp, role, content).
• remo-conversation-context: Stores conversation context (agent state, history, keywords, etc.).
• remo-users: Stores user details.

Each table is managed in src/utils/dynamodb_service.py.

---

How AWS Bedrock is Used

• The main AI models (for chat, reasoning, and image generation) are run on AWS Bedrock.
• Each agent (and the supervisor) can use Bedrock models for smart responses.
• ContentCreatorAgent uses Bedrock to generate images.
• See: src/agents/content_creator/content_creator_tools.py and agent files.

---

How Google Client is Used

• For scheduling meetings and calendar events (mainly in EmailAgent).
• Uses Google Calendar API via helper functions.
• See: src/utils/google_calendar_service.py and src/agents/email/email_tools.py.

---

Message Flow (Like a Story)

1. You say: "Remind me to call mom at 5pm."
2. Remo (Supervisor) thinks: "This is a reminder! I'll ask the ReminderAgent."
3. ReminderAgent says: "Okay, I'll set a reminder for 5pm to call mom."
4. ReminderAgent uses its tool: Calls set_reminder to save the reminder in DynamoDB.
5. Remo gets the answer: "I've set a reminder to call mom at 5pm for you!"
6. You see the answer.

---

All the Possibilities (What You Can Ask)

• Reminders: Set, list, update, delete, mark as complete.
• Todos: Add, list, update, delete, mark as complete, prioritize.
• Emails: Compose, send, schedule, search, mark as read, archive, forward, reply, summarize, schedule meetings.
• Images: Generate images from text prompts.

---

How to Add More

• New Agent: Create a new agent class, add tools, register in SupervisorOrchestrator, add routing.
• New Tool: Add a Python function, wrap it for the agent, and document what it does.

---

Where to Find Everything

• Supervisor: src/orchestration/supervisor.py
• Agents: src/agents/
• Agent Tools: src/agents/*_tools.py
• DynamoDB Service: src/utils/dynamodb_service.py
• Google Calendar: src/utils/google_calendar_service.py
• Content Creation: src/agents/content_creator/content_creator_tools.py

---

For Developers (and Curious Users)

• The system is modular and easy to extend.
• All user requests go through the Supervisor, which delegates to agents.
• Agents use LangGraph and LangChain for smart reasoning and tool use.
• Data is stored in DynamoDB tables, organized by user and type.
• AWS Bedrock powers the AI models.
• Google client is used for calendar integration.

---

If you have any questions, just ask Remo or check the code in the folders above. This README is your map to the Remo universe!