dzhng/deep-research

Open Deep Research is an AI-powered research assistant that automatically investigates any topic by combining web searches, content extraction, and large language models (AI text systems like GPT). The problem it solves: researching a complex topic thoroughly requires multiple rounds of searching, reading, connecting ideas, and generating new questions, a tedious process when done manually. This tool automates that loop. Here's how it works: you enter a research question and set two parameters, "breadth" (how many parallel search threads to pursue) and "depth" (how many levels of follow-up exploration to do). The system generates targeted search queries using an AI model, searches the web via Firecrawl (a web scraping service), extracts the key learnings from the results, identifies new research directions those findings suggest, and then recursively searches those directions. After completing all iterations, it compiles everything into a comprehensive Markdown report with sources. You would use this when you need to deeply understand a topic quickly, competitive research, technical investigation, academic background reading, and want the AI to do the iterative "what should I read next" work for you. It requires a Firecrawl API key for web search and an OpenAI API key (for the o3-mini model) or can use local AI models via OpenAI-compatible endpoints. The codebase is intentionally kept under 500 lines of TypeScript so it's easy to understand and build on.