Can I vibecode 100 Questions?

Build a robust, local AI visibility and benchmarking tool designed to replace the core functionality of the "100 Questions" app. This tool will systematically evaluate how well a specific brand is represented across multiple AI models and search providers, offering deep insights without recurring SaaS fees.
Use this stack: Node.js 22, TypeScript, SQLite for local data storage, and the official SDKs for OpenAI, Anthropic, Google Gemini, and xAI. The interface should be a powerful command-line interface (CLI) coupled with a static HTML report generator.
The core workflow: The user runs `benchmark --domain [brand.com] --description "[brand description]"`. The system must generate 100 diverse buyer and industry questions based on the provided domain and description using an LLM, or accept a JSON file of pre-defined questions.
For every question, the system must query all four providers (OpenAI, Anthropic, Gemini, xAI) simultaneously. It must utilize each provider's native web-search or grounding tools to ensure the models have access to the latest internet data. 
All API keys must be securely loaded from a local `.env` file. Do not hardcode any credentials.
Implement robust error handling: limit concurrency to respect rate limits, automatically retry transient network or API failures using exponential backoff, and ensure that permanent failures are logged and preserved in the final report rather than crashing the run.
Store every piece of data in SQLite: the exact prompt used, the raw text response from each model, extracted citation URLs, timestamps, model IDs, and any encountered errors.
After collecting responses, run a secondary structured LLM pass over the data to detect brand mentions (including case-insensitive and alias matches configured in a local file) and extract named competitors mentioned in the answers.
Normalize all citation URLs by parsing the hostname, resolving canonical URLs, and stripping out UTM or tracking parameters to provide a clean list of sources.
Calculate detailed metrics: brand visibility percentage per provider, overall answer coverage, the rate of citations pointing to the owned domain, and a ranked list of the top overall sources cited by the AIs.
Generate a polished, self-contained static HTML report. The report must include interactive filters, data visualizations for the metrics, and the ability to expand and read the raw evidence (the exact LLM response) for every question.
Include a feature to export the raw data (questions, metrics, competitors, citations) to CSV files for further analysis.
Do not include user accounts, billing, scheduled monitoring, or a web dashboard. Keep this strictly as a local CLI tool.
Write fixture-based unit tests for the URL normalization logic, brand mention detection, and metric calculations. Provide a comprehensive README detailing setup, API key requirements, grounding caveats per provider, and examples of exact run commands.