This video provides a rapid tour of the Solveit platform, a novel development environment presented by Jeremy Howard and Jono. The platform is designed as a hybrid workspace that integrates the features of a Jupyter notebook, an AI chatbot, and a command-line shell into a collaborative interface called a "dialogue." Users operate within personal virtual servers known as "instances," where they can write and execute code, take notes, and interact with an AI.
Key features demonstrated include different AI interaction modes (Learning, Standard, Concise) to suit various user needs, and a "tools" system that allows the AI to execute user-defined Python functions for agent-like automation. The platform supports real-time collaboration, live web application hosting via public URLs, and the ability for the AI to analyze screenshots of the UI for visual tasks. The overarching philosophy encourages an iterative, "human-in-the-loop" workflow, with functionalities for exporting work into reusable Python modules, compatible with tools like nbdev.
A Guided Tour of the Solveit Platform's Features and Philosophy
0:00:32 Instances as Personal Servers: Each user gets one or more "instances," which function as isolated, personal virtual private servers. These instances have their own file systems and can be used to host applications and manage files.
0:03:29 Dialogues: The Core Workspace: The central concept is the "dialogue," a saved document that combines notes (Markdown), executable code (like a Jupyter notebook), and prompts for an AI (like a chatbot). This allows users to maintain a complete, interactive record of their work.
0:06:12 Real-Time Collaboration: Users can share a private URL to their dialogue, allowing multiple people to view and edit the content simultaneously, making it a collaborative tool by default.
0:10:47 The Solver Method Philosophy: The platform's design encourages an iterative workflow with small, manageable steps and quick feedback, discouraging large, monolithic blocks of code.
0:29:51 Tailored AI Interaction Modes: Solveit offers different modes to control the AI's behavior:
Learning Mode: The AI asks clarifying questions to understand the user's knowledge level before providing answers, making it ideal for learning.
Concise Mode: Provides direct, compact answers and code, suited for experienced users who want to work quickly.
Standard Mode: Offers a default, more verbose interaction style similar to standard chatbots.
0:37:00 Agent-like AI with Tools: Users can make Python functions available to the AI as "tools." This allows the AI to perform complex, multi-step tasks such as searching websites, reading files, and running code, functioning like a customizable AI agent.
0:38:17 Smart Code Completion: The platform features "ghost text" (AI-powered code suggestions), which is intentionally disabled in Learning Mode to encourage users to think through the code themselves.
0:55:30 Live Web App Hosting: Users can run web servers (e.g., FastHTML, FastAPI) on their instance, which become accessible to the public via a unique URL, enabling rapid prototyping and deployment of web applications.
1:07:19 AI Vision for UI Tasks: The platform allows the AI to take a screenshot of the rendered output (e.g., an HTML table). This enables the AI to "see" what the user sees and provide more accurate assistance with visual tasks like CSS styling.
1:11:24 Exporting Code for Reuse: Code cells marked for export can be turned into a standard Python module (.py file) with a single click. This integrates with development workflows like nbdev, allowing work done in Solveit to be easily incorporated into larger projects.