hoolulu/deep-research

What it solves

This project provides an automated system for generating institutional-grade deep research reports. It addresses the common limitations of AI research, such as shallow summaries, high costs of professional reports, difficulty accessing specific regional sources (like Chinese platforms), and the tendency for AI to fabricate data without traceable sources.

How it works

The system uses a multi-agent autonomous pipeline consisting of four stages:

1. Analyze Outline: The AI analyzes the topic and creates a research framework and search plan.
2. Collect Data: It gathers information using a four-layer priority search strategy (built-in engines, suggested sources, a curated sources.json list, and free fallbacks) and fetches content via Scrapling. It also supports offline mode by reading local PDF, DOCX, TXT, and MD files.
3. Parallel Writing: Chapters are written in parallel, embedding facts directly into the prompts to avoid tool calls.
4. Validate & Assemble: The system performs batch validation, assembles the report, converts citations, and runs a final QA check.

Who it’s for

It is designed for indie developers, independent researchers, and small teams who need professional-grade research capabilities without the expense of paid databases or research institutions.

Highlights

• Multilingual Support: Auto-detects the topic language and generates reports in 19 different languages natively.
• Traceable Sources: Every number and fact is backed by clickable citations and a full reference list.
• Comprehensive Analysis: Includes opposing viewpoints in every chapter and provides confidence grading for the final summary.
• Flexible Depth: Offers three modes (Quick, Standard, and Deep) to control report length and detail.
• ** sublattice Tool Integration**: Works across various AI coding tools like Claude Code, Cursor, and Codex CLI.
• Local Management: Includes a local report browser for searching, filtering, and exporting reports to PDF or DOCX.