This document is a transcript of conversations from the "new-solvers" Discord channel for the SolveIt platform, spanning from October 13 to October 18, 2025. The discussions capture a wide range of user interactions, from beginners asking fundamental questions to experienced users exploring advanced functionalities and workarounds.
Key topics include core platform usage (file management, UI navigation, keyboard shortcuts), advanced techniques like multi-language support (Mathematica), connecting to local machines via SSH tunnels, and developing web applications using FastHTML. Users troubleshoot common issues such as library incompatibilities (JupyUvi, Playwright), AI hallucinations, and backend API errors. The transcript also reveals insights into the platform's philosophy, future pricing models (around $10/month with no self-hosting), and the community-driven approach to documentation and support, with the platform's creator and TAs actively participating in the dialogue.
Key Discussions and User Questions in the SolveIt 'new-solvers' Channel
Oct 13, 4:53 PM - Multi-Language Support: A new user asks about using languages other than Python (e.g., Typescript, Java).
Key Takeaway: While not officially supported, the community is exploring workarounds. A TA later demonstrates running Mathematica commands directly within a SolveIt dialog, showing that integration with other languages is possible.
Oct 13, 7:05 PM - Clearing Cell Outputs: A user requests a feature to clear all cell outputs, similar to Jupyter notebooks, to ensure outputs are from the latest run.
Key Takeaway: Platform creator Jeremy Howard explains that restarting a dialog clears the execution timestamps on cells, which serves the same purpose of indicating which outputs are fresh.
Oct 13, 7:17 PM - Keyboard Shortcut Confusion: A user discussion highlights that the keyboard shortcuts for deleting (x) and duplicating (Shift+D) cells differ from Jupyter's conventions, causing confusion and accidental dialog duplication.
Oct 14, 9:31 AM - Folder Creation: A user asks how to create folders to organize their work.
Key Takeaway: Folders are created automatically by including a forward slash (/) in a dialog's name (e.g., my_project/data_analysis). The mkdir command in the built-in terminal also works.
Oct 15, 11:56 AM - Project Organization: A user asks for best practices on when to create multiple notebooks (dialogs) versus new instances.
Key Takeaway: The recommended practice is to use multiple dialogs within a single instance for related work, organized with folders. New instances should only be created for projects requiring complete environment isolation.
Oct 15, 3:46 PM - Connecting to Local Machines: Users discuss and successfully implement ways to connect their cloud-based SolveIt instance to their local computer.
Key Takeaway: Using tools like ngrok to create a secure SSH tunnel allows users to run code (e.g., Python scripts, GPU tasks) on their local machine and pipe the output back into SolveIt for analysis.
Oct 15, 8:55 PM - FastHTML App Development Issues: A user reports that their SolveIt instance becomes unresponsive when trying to run a FastHTML web app.
Key Takeaway: The JupyUvi library is designed for Jupyter and will crash a SolveIt instance. The correct function for serving web apps within SolveIt is nb_serve_async.
Oct 16, 12:02 PM - Referencing Other Dialogs: A user asks how to reference notes from one dialog while working in another without extensive copy-pasting.
Key Takeaway: The most effective method is to write a Python function that reads the content of the reference dialog and then register that function as a custom tool for the LLM to use on demand.
Oct 16, 7:39 PM - Unsupported Libraries: A user trying to follow a web scraping tutorial runs into dependency issues with the Playwright library.
Key Takeaway: Playwright is not currently supported. Jeremy Howard suggests using Browserbase as a viable alternative for browser automation tasks.
Yesterday at 2:38 AM - Platform Errors (E300): A user reports receiving an "E300" error, preventing them from using the platform.
Key Takeaway: The E300 error code indicates that the backend LLM provider (Anthropic's Claude API) is overloaded or timing out, an issue that can occur even when the official status page reports no problems.
Saturday, 6:43 AM - Handling AI Hallucinations: A user describes an instance where the AI invented a non-existent function (stop_render_ft()).
Key Takeaway: When the AI hallucinates, users can provide it with more context (e.g., by running dir() on the relevant object) or, more productively, ask the AI to help write the missing function, turning the error into a learning opportunity.