Claude Code theme

Deep Researcher

Deep Research based on Claude Code's agentic framework export const SPINNER_VERBS = [ 'Accomplishing', 'Actioning', 'Actualizing', 'Architecting', 'Baking', 'Beaming', "Beboppin'", 'Befuddling', 'Billowing', 'Blanching', 'Bloviating', 'Boogieing', 'Boondoggling', ...]

106 stars 36 forksThemes & HUDs Source

Install

From source

git clone https://github.com/jackswl/deep-researcher.git

From the README

&nbsp;&nbsp;Deep Researcher

  Search Google Scholar, enrich with OpenAlex, and extract a structured literature matrix, all locally.

  Quick Start &middot;
  How It Works &middot;
  Sample Output &middot;
  Configuration

Deep Researcher searches Google Scholar for academically-ranked papers, enriches them with full metadata from OpenAlex and CrossRef, and uses a local LLM to extract a structured matrix of each paper (method and key finding), grouped by theme.

- 100 papers from Google Scholar's semantic search, no keyword hacks, no irrelevant results
- Full metadata: DOIs, abstracts, journal names, citation counts, open access URLs
- Structured extraction: papers grouped by theme, each reduced to the method and key finding stated in its abstract
- Consistent [N] references: every row in the tables matches the reference list
- BibTeX + CSV output ready for LaTeX/Overleaf or Excel
- Runs locally with Ollama. Your queries never leave your machine
- Tool-based agentic architecture inspired by Claude Code
- 4 dependencies, no LangChain

Quick Start

Run with Ollama (local, free, private)

Run with a cloud provider

All supported providers

How It Works

1. Search: Queries Google Scholar for up to 100 semantically-ranked academic papers
2. Enrich: Concurrent workers (8 threads) look up each paper in OpenAlex/CrossRef for full abstracts, DOIs, and journal metadata
3. Extract: LLM groups papers into themes, then extracts a table per theme (method and key finding for each paper, taken only from the abstract)

Each run produces:

Sample Output

First theme shown in full, remaining themes truncated for brevity
Every [N] in the tables matches [N] in the reference list. Every row is extracted from a real abstract: when an abstract does not state the method or finding, the cell…

Read the full README on GitHub