Summary
Highlights
The video introduces Federico Costa, an AI expert, who will guide viewers through creating a 24/7 AI assistant named Jarvis, accessible via Telegram. Jarvis, inspired by Iron Man's AI, aims to automate operational tasks like managing clients, generating quotes, and handling emails, freeing users from constant computer reliance. This approach contrasts with existing solutions like OpenClow, which can be costly and less secure due to token-based payments and community-driven development. The proposed solution uses Cloud Code with a subscription model, offering a fixed cost and more streamlined, specific functionalities tailored to individual needs.
Jarvis is designed to be personalized, knowing user preferences and business details. It interacts via Telegram, a preferred platform over WhatsApp due to its simpler integration process. Jarvis uses a database for storing client and project information and can perform tasks like setting reminders, drafting quotes, and providing daily briefings. The main advantage of this setup, compared to running AI on a personal computer, is continuous availability, lower energy costs, and immunity to system updates or standby modes, achieved by hosting Jarvis on a Virtual Private Server (VPS).
The first step is to acquire a VPS, a virtual computer hosted on a data center for a low monthly fee (around €5-€10). Hostinger is recommended for its user-friendliness for non-technical users. The process involves selecting a plan, choosing a server location (preferably near the user for lower latency), and selecting an operating system like Ubuntu LTS. A root password is set for administrative access, and optional features like Docker Manager can be pre-installed. The video also explains Docker, a system that packages applications with their dependencies, creating isolated 'containers' to ensure stability and security.
The video provides a pre-configured Jarvis template (available for download) to streamline the setup. This template includes a structured workspace for Jarvis to manage information (e.g., 'about me', database, skills) and includes a 'cloud_install' file that guides Cloud Code through the installation process. The workflow involves the VPS hosting Docker containers, which house Cloud Code (the AI agent), a Telegram plugin, a database, and integrations with third-party services. Docker volumes store files externally for backup, and GitHub is used for version control and synchronization. The connection to the VPS is established via SSH using a tool like Warp, which simplifies terminal commands with AI assistance.
Cloud Code prompts the user for personal information (name, business type) to customize Jarvis's context. This personalization helps Jarvis respond more accurately and efficiently. The importance of the '.ENV' file for securely storing sensitive information like API keys and passwords is highlighted. Next, the process of creating a Telegram bot using BotFather is detailed, where a unique token is generated for Jarvis to communicate via Telegram. This token is then securely passed to Jarvis. The system automatically creates a GitHub repository to track changes and updates, including user-defined preferences and new functionalities.
To enable Jarvis to transcribe voice messages, an Open AI API key is generated and securely added to Jarvis. This leverages Open AI's Whisper model, known for its cost-effectiveness. A credit card linked to Open AI is required for this service. The video then demonstrates integrating an email service using Agent Mail, a platform designed for AI agents. An email address is created for Jarvis, and its API key is provided, allowing Jarvis to send and receive emails. The possibility of setting up webhooks for real-time email notifications is mentioned, though a simpler manual check is used for the demo.
A significant customization step is creating a Telegram Web Application (TWA) directly within Jarvis. This TWA can display tasks, daily briefings, and graphical reports. A domain with SSL certification is required to host the TWA securely. Cloudflare is used to manage the domain records, ensuring proper routing and security for the TWA. Cloud Code assists in designing the TWA interface, demonstrating how to integrate features like task management with read/write access. This showcases the potential for building mini-applications within Telegram for enhanced productivity.
The AI assistant actively debugs and fixes issues during deployment, such as permission problems or configuration errors, highlighting the AI's ability to self-correct. The power of parallel processing for managing multiple AI agents and projects concurrently is demonstrated using Warp. After successful deployment, Jarvis shows its fully integrated capabilities, including sending emails and responding to requests, even with personalized greetings. The video concludes by emphasizing that this setup allows for continuous operation regardless of the user's computer status, making it a highly flexible and powerful tool for automation and personal assistance.
The video summarizes the achievements: setting up an advanced AI assistant, configuring a VPS, integrating Telegram and email, and building a custom TWA. Future possibilities include connecting Jarvis to services like Eleven Labs for voice responses or phone numbers for automated calls. The template's expandable nature is highlighted, encouraging viewers to suggest new functionalities. The hosts reiterate the efficiency and cost-effectiveness of this setup, especially compared to token-based models. The video concludes with a small demonstration of the working TWA and an invitation for community feedback on future video topics.