Comparative Analysis

Comparative Analysis: DeepDelve versus LLM Wiki and Deep Research Agent

Compares DeepDelve against two reference implementations, nashsu/llm_wiki (persistent knowledge base and wiki paradigm) and CYC2002tommy/Deep-Research-Agent (academic pipeline, "Deep Science Writer"), looking for concrete features worth adopting.

Feature matrix

Feature / Dimension	DeepDelve	LLM Wiki	Deep Research Agent
Primary Goal	General purpose web research	Incremental local wiki	Academic literature review
Output Format	Markdown (final_report.md)	Interlinked Obsidian pages	APA .docx manuscript
Search Engine	DuckDuckGo by default, Tavily/Brave optional, snippets + top 1 fetch	Keyword + vector hybrid on the wiki	OpenAlex, Scopus, Semantic Scholar, Exa
Knowledge State	Ephemeral, per run session folders	Persistent, living wiki	Ephemeral, one study draft
Verification Gate	URL presence + term overlap	Review queue, user managed	Strict DOI lookup, journal rank, peer critique

Versus llm_wiki

DeepDelve is ephemeral RAG, every query from scratch into a timestamped folder. llm_wiki treats the agent as a librarian maintaining a living, persistent knowledge graph instead. Worth adopting: a wiki mode where the workspace persists across runs (concepts/, sources/, a living index.md), automatic [[Page Name]] cross linking so Obsidian resolves a visual graph, and control files (purpose.md, schema.md) the Planner reads at startup to keep naming and structure consistent.

Versus CYC2002tommy/Deep-Research-Agent

Note

AcademicSearcher now has a real, optional Tavily or Brave search backend, not DuckDuckGo only. That's still general web search, though, not scholarly metadata; the actual gap this section describes (no OpenAlex/DOI/journal-rank layer) is still real and unimplemented.

DeepDelve's AcademicSearcher pulls general web content instead of scholarly metadata, whichever search backend it's using. CYC2002tommy integrates OpenAlex, Semantic Scholar, and Scopus directly, with strict DOI validation and journal rank screening. Worth adopting: an academic_search tool against the free OpenAlex API, journal quality filtering to avoid predatory or low tier sources, a verify_doi tool against CrossRef, and elevating the existing "Global Critic" pass into a real peer review simulation checking for sample size limitations, overgeneralization, and conflicts of interest.

A rough implementation sketch

Phase A, the academic toolkit: academic_query_openalex(query, limit) returning structured title/DOI/authors/journal/quartile/abstract JSON, plus verify_doi(doi) against CrossRef. Phase B, the wiki module: list_wiki_links() so the Planner knows what to cross reference, and a settings.workspace.mode: "wiki" option that bypasses session isolation in favor of one consolidated, interlinked workspace.