What problem does it solve? Turning a pile of local PDFs, notes, and web clippings into a structured, interlinked research wiki is slow and error-prone, especially when discovery, deduplication, and multi-paper ingestion must stay consistent. ## Core Features & Use Cases - Deterministic source preparation: Normalizes local PDFs into prepared sources under raw/tmp/ with strict title and arXiv ID recovery rules, recording warnings instead of aborting. - Planner-guided discovery: Builds a shortlist via Semantic Scholar, DeepXiv, and arXiv, then requires an explicit LLM trim to a final 8-10 paper set before fetching into raw/discovered/. - Parallel ingest with worktree isolation: Fans out one /ingest subagent per paper in git worktrees, then merges sequentially, deduplicates graph edges and citations, and rebuilds indexes. - Use Case: A researcher drops five PDFs into raw/papers/ plus notes in raw/notes/, runs /init with a topic, and receives a scaffolded wiki with provisional pages, discovered papers, and fully ingested paper pages with a checkpointed, resumable manifest. ## Quick Start Run /init with your research topic after placing your papers, notes, and web sources under the raw/ directory to bootstrap the wiki and ingest the final paper set in parallel.
