answer_pdf_queries
Answers a question about one or more PDFs given their URLs. Pass multiple URLs to compare or synthesize across papers. Supports arXiv, alphaXiv, and semantic scholar abstract pages. Remember that calling one query with multiple papers and calling multiple queries with one paper each are both valid strategies to use this tool.
get_paper_content
Get the content of an arXiv/alphaXiv paper as text. By default returns a structured AI-generated intermediate report (optimized for LLM consumption). If no report is available, automatically falls back to the full extracted text. Use the fullText option to skip the report and get the raw extracted text directly.
read_files_from_github_repository
Reads the contents of a file or directory from the paper's codebase repository. Special behaviors:
1. When path is '/', returns a complete file tree AND the contents of all top-level files in the repository
2. When path points to a directory, returns a list of files and directories, with contents of all files fetched in parallel. Use if you want to fetch multiple files at once. Note: you are encouraged to read files this way rather than reading individual files which can be time-consuming.
3. When path points to a file, returns its contents
embedding_similarity_search
Search for papers by semantic/conceptual similarity using embeddings. Returns top papers ranked by similarity and popularity. Best for finding papers about specific concepts, methods, or research areas. Call multiple times with varied queries to cover different angles if initial results are incomplete.
full_text_papers_search
Search the alphaXiv database for arXiv papers by keyword. Best for keywords, method names, benchmarks, or author names. Iterate with different keyword combinations if initial results don't fully answer the question.
agentic_paper_retrieval
An additional retrieval tool that autonomously performs multi-turn searches to find relevant academic papers. IMPORTANT: Always call this IN PARALLEL with `embedding_similarity_search` and `full_text_papers_search`, never instead of them. Each tool covers different blind spots — this tool provides extra recall but does not replace the others. Best for grounded-research queries where comprehensive coverage matters.