perspectv

Creates an LLM-powered summary of a business from its website.

💭 Introduction

perspectv is a command-line tool that scrapes a business website and creates a report. The content of the report is currently:

• Business summary
• Product summary
• SWOT
• Software & data strategy

LLMs are used to extract content from the pages of the company's website, and to generate the report. Two different models may be used (e.g. a low-cost extraction model and a large-context reporting model -- since the latter model is expected to take the entire website content as context).

🚀 Running

While perspectv is in beta, it must be run from the repo directory using poetry:

poetry run perspectv example.com

A single environment variable is required, OPENROUTER_API_KEY. Currently, (OpenRouter)[https://openrouter.ai/] is used to provide a strong model selection. In the future, it would be beneficial to support other LLM vendors directly.

🧂 Options

• --dbfile: Filename for a SQLite database used to store incremental data for the web scrape and the LLM usage. perspectv will by default avoid redoing the same work. Deleting the database or using a new file will start over.

• --model-extract: Override the OpenRouter model name used for web page extracts. This should be a low-cost model with reasonable extraction performance (many models can do this acceptably).

• --model-report: Override the OpenRouter model name used for report generation. This shoudl be a large-context model. The default, anthropic/claude-3-opus, is relatively expensive but performs well and has a 1 million token context window.