advanced_search
ChatGPTUse this for complex OpenAlex searches with multiple field-specific criteria combined with boolean operators. What it does: - Constructs sophisticated multi-field queries for OpenAlex - Combines multiple search criteria with AND/OR logic - Searches specific fields (title, author, abstract, institution, etc.) - Provides maximum precision for complex research questions When to use: - User wants papers matching MULTIPLE specific criteria - Need to search different fields with different terms - Complex queries like "papers about X by author Y from institution Z" - User wants to combine title search + author search + date filter - Systematic reviews requiring precise inclusion criteria How to use: - query_parts: Array of {field, query} objects Examples: [{field: "title", query: "neural networks"}, {field: "author.display_name", query: "LeCun"}] [{field: "abstract", query: "deep learning"}, {field: "authorships.institutions.display_name", query: "MIT"}] - boolean_operator: "AND" (all must match) or "OR" (any can match) - filters: Additional filters like publication date - sort: Sorting preference - per_page/page: Pagination Returns: Same as search_open_access_works - list of matching papers. Example: Find papers with "climate change" in title AND "machine learning" in abstract from 2023+: query_parts=[{field:"title", query:"climate change"}, {field:"abstract", query:"machine learning"}] boolean_operator="AND" filters={"from_publication_date": "2023-01-01"}
arxiv_search
ChatGPT⚠️ RARELY NEEDED - Use search_open_access_works with arXiv filter instead! PREFERRED ALTERNATIVE: search_open_access_works(query="your keywords", filters={"locations.source.display_name": "arXiv"}) This gives better results with OpenAlex metadata! What it does: - Direct search of arXiv preprint repository - Covers physics, math, CS, and related fields - Returns arXiv-specific IDs and metadata When to use (RARE): - User explicitly says "search arXiv directly" or "use arXiv API" - search_open_access_works with arXiv filter fails - Need arXiv-specific XML metadata When NOT to use: - User wants arXiv papers → use search_open_access_works with arXiv filter (better!) - General physics/CS/math search → use search_open_access_works Parameters: - query: Search terms - start: Pagination index (default 0) - max_results: Number of results (default 10) Returns: {status, data: [{id, title, summary, authors, published, pdf_url}]} 🎯 WORKFLOW: User wants arXiv → search_open_access_works + arXiv filter (preferred)
arxiv_work_info
ChatGPT⚠️ LAST RESORT - Use ONLY if both get_open_access_work AND url_extract fail. PREFERRED ALTERNATIVES (in order): 1. get_open_access_work (use DOI/OpenAlex ID from search) - BEST 2. url_extract (use arXiv URL) - SECOND CHOICE 3. arxiv_work_info (this tool) - LAST RESORT ONLY What it does: - Fetches paper from arXiv by arXiv ID - Has rate limits When to use (RARE): - get_open_access_work failed - url_extract failed on arXiv URL - User provides arXiv ID specifically Parameters: - arxiv_id: arXiv ID like "2301.12345" Returns: {status, data: {id, title, summary, authors, published, pdf_url}} 🎯 WORKFLOW: 1. get_open_access_work → 2. url_extract → 3. arxiv_work_info (last resort)
biorxiv_search
ChatGPT⚠️ RARELY NEEDED - Use search_open_access_works with bioRxiv filter instead! PREFERRED ALTERNATIVE: search_open_access_works(query="your keywords", filters={"locations.source.display_name": "bioRxiv"}) Better results with OpenAlex metadata! What it does: - Direct search of bioRxiv preprint server - Title-only matching (limited) - Has rate limits When to use (RARE): - User explicitly says "search bioRxiv directly" - search_open_access_works with bioRxiv filter fails When NOT to use: - User wants bioRxiv papers → use search_open_access_works with bioRxiv filter (better!) Parameters: - query: Search terms (title matching) - year_start/year_end: Year range - sorted: true for date sort - count: Results (default 20) Returns: {query, results: [{paper metadata}]} 🎯 WORKFLOW: User wants bioRxiv → search_open_access_works + bioRxiv filter (preferred)
biorxiv_work_info
ChatGPT⚠️ LAST RESORT - Use ONLY if both get_open_access_work AND url_extract fail. PREFERRED ALTERNATIVES (in order): 1. get_open_access_work (use DOI from search) - BEST 2. url_extract (use bioRxiv URL) - SECOND CHOICE 3. biorxiv_work_info (this tool) - LAST RESORT ONLY What it does: - Fetches paper from bioRxiv API - Has rate limits When to use (RARE): - get_open_access_work failed - url_extract failed on bioRxiv URL Parameters: - biorxiv_id: bioRxiv DOI or title - id_type: "doi" or "title" Returns: {identifier, id_type, paper: {metadata}} 🎯 WORKFLOW: 1. get_open_access_work → 2. url_extract → 3. biorxiv_work_info (last resort)
fetch
ChatGPTInternal APP fetch tool for accessing internal APIs
files_rag_answer
ChatGPT⚡ TARGETED FILE RETRIEVAL TOOL - Use when you have specific file IDs ⚡ PRIMARY USE CASE: When file IDs are explicitly provided or known (e.g., from user context, previous searches, or document references). ✅ WHEN TO USE THIS TOOL: - File IDs are specified in the user's request or context - You need to retrieve information from SPECIFIC documents (not the entire knowledge base) - User asks to analyze/summarize/query PARTICULAR files - Prompts explicitly mention using "files_rag_answer" tool - Working with a targeted subset of documents - Following up on specific papers/documents from previous operations ❌ WHEN NOT TO USE (use wisebase_rag_answer instead): - No specific file IDs are available - Need to search across the ENTIRE knowledge base - Exploratory search without file constraints - Multi-round iterative retrieval needed 🎯 ADVANTAGES: - Direct, focused retrieval from specified files - Faster response time (no multi-round iteration) - Precise control over search scope - Better for specific document analysis INPUT REQUIREMENTS: - fileIds: Array of target file IDs (REQUIRED - must have at least one) - question: Your search query or question CRITICAL: If you have file IDs available in the context, YOU MUST use this tool instead of wisebase_rag_answer.
get_open_access_work
ChatGPT🎯 SCHOLAR'S PAPER LOOKUP TOOL - Use this for getting detailed information about a single paper. Superior to all platform-specific fulltext tools. What it does: - Retrieves complete metadata for a single paper from OpenAlex (250M+ papers indexed) - Accepts DOI, PMID, or OpenAlex ID from previous search results - Returns comprehensive details: all authors, institutions, citations, references, related works, PDF links - Cheapest option for single paper lookups, no rate limits, better data quality than platform-specific tools ⭐ HIGHEST PRIORITY for single paper lookups! When to use: - You have a paper identifier (DOI, PMID, OpenAlex ID) from a previous search result - User says "tell me more about this paper" (use ID from context) - User wants detailed information about a specific paper - Need full author list, institution details, or citation data for one paper - ALWAYS use this instead of pubmed_work_info/arxiv_work_info/biorxiv_work_info When NOT to use: - User doesn't have a paper ID yet → use search_open_access_works first - Searching for papers (not looking up a specific one) → use search tools How to use: - identifier: Usually comes from previous search results (results[0]["id"] or results[0]["doi"]) OpenAlex ID: "W2124808589" (from search results) DOI: "10.1038/s41586-023-12345-6" (from search results) PMID: "36878453" (from search results) Returns: Complete paper object with all metadata, authors, institutions, citations, referenced_works, related_works, credibility indicators. 🎯 WORKFLOW: 1. Search finds papers → results contain IDs 2. Use get_open_access_work with that ID for full details 3. NEVER use pubmed_work_info/arxiv_work_info/biorxiv_work_info if this works 💰 COST:* Cheapest option for single paper details - No rate limits, superior data.
knowledge_graph
ChatGPTGenerate a non-hierarchical, network-based knowledge graph visualization.
knowledge_graph_list
ChatGPTdisplay use's knowledge graph list.
prepare_chat_for_saving
ChatGPTWhen a user wants to create a document that they intend to save for future use, this tool should be invoked to analyze and generate the document according to the user's instructions; after the user confirms, the save function will then be called. Otherwise, the save function should not be invoked without the user's permission.
pubmed_search
ChatGPT⚠️ RARELY NEEDED - Use search_open_access_works with PubMed filter instead! Only use this for NCBI-specific databases beyond PubMed. PREFERRED ALTERNATIVE: search_open_access_works(query="your keywords", filters={"locations.source.display_name": "PubMed"}) This gives better results and includes OpenAlex metadata! What it does: - Direct search of NCBI databases (PubMed, PMC, Gene, Protein, etc.) - Returns PubMed-specific summaries with PMIDs - Provides MeSH terms and NCBI-specific metadata When to use (RARE): - User explicitly asks for "NCBI Gene database" or "NCBI Protein database" (non-PubMed NCBI) - Need MeSH terms specifically - User wants to search PMC, Gene, or Protein databases - search_open_access_works fails after trying with PubMed filter When NOT to use: - General medical/biological paper searches → use search_open_access_works - User says "find PubMed papers" → use search_open_access_works with PubMed filter (better!) Parameters: - query: Search query - database: "pubmed" (default), "pmc", "gene", "protein" - max_results: Default 10 Returns: {search_results: {count, query_key, webenv}, summaries: {pmid: {metadata}}} 🎯 WORKFLOW: 1. User wants PubMed → search_open_access_works with PubMed filter 2. Only use pubmed_search if they need Gene/Protein databases
pubmed_work_info
ChatGPT⚠️ LAST RESORT - Use ONLY if both get_open_access_work AND url_extract fail. PREFERRED ALTERNATIVES (in order): 1. get_open_access_work (use paper ID from search results) - BEST 2. url_extract (use PubMed URL) - SECOND CHOICE 3. pubmed_work_info (this tool) - LAST RESORT ONLY What it does: - Fetches paper info directly from PubMed/PMC via NCBI APIs - Has rate limits and slower than get_open_access_work - Returns basic metadata: title, abstract, DOI, journal, dates When to use (RARE): - get_open_access_work failed (paper not in OpenAlex) - url_extract failed on PubMed URL - User provides only a PMID and get_open_access_work didn't work How to use: - identifier: PMID, PMC ID, DOI, or paper title - id_type: "pmid", "pmc", "doi", or "title" Returns: {title, abstract, pdf_link, full_text_link, journal, pub_date, issn} 🎯 CRITICAL WORKFLOW: 1. Have paper ID → get_open_access_work 2. If fails → url_extract with PubMed URL 3. If still fails → pubmed_work_info (this tool)
save_edited_document
ChatGPTThe model must not actively invoke this tool during the conversation. This tool is only intended to be triggered by the frontend via window.openai.calltool after chat-save-preparation has been called, and it is used solely for saving files that have already been edited. Therefore: - The model must not call this tool directly in its reasoning or responses. - The timing and logic for invoking this tool are entirely controlled by the frontend, and are independent of any tool calls produced by the model.
scholar_search
ChatGPT🎯 Use this for all Google Scholar academic searches. What it does: - Searches Google Scholar database (includes papers from all publishers, indexed content) - Returns papers with titles, snippets, publication info, citation counts, and links - Includes both open-access and paywalled papers (shows access status) - Fast and reliable for academic literature searches When to use: - User wants scholarly articles from Google Scholar - Need to check citation counts for papers - User asks for papers by specific authors - Looking for well-known, highly-cited works - Literature reviews and research tasks - User asks "find papers on X" or "search for research about Y" Parameters: - query: Search terms (required) - language: Language code (default "en", supports "zh-CN" for Chinese) - year_start/year_end: Filter by publication year range - sorted: false for relevance (default), true for date - count: Number of results (default 5, recommend 20+ for comprehensive searches) Returns: List of papers with title, authors, snippet, publication_info, citation count, links, and resource type (PDF/HTML/Book). ⭐ RECOMMENDED: Use count=20 or higher for comprehensive search results.
search_open_access_works
ChatGPT🎯 SCHOLAR'S PRIMARY SEARCH TOOL - Use this FIRST for all academic searches. Cheapest option, fast, and covers 250M+ papers from Scholar's OpenAlex integration. What it does: - Direct search of OpenAlex database (world's largest open catalog of scholarly works) - Works best with simple keyword searches (2-5 keywords) - Supports optional simple filters (mostly used WITHOUT filters) - Allows sorting by date, citations, or relevance - Returns comprehensive paper metadata with DOIs, PDFs, citations, authors, institutions ⭐ DEFAULT CHOICE: Use this for 90% of academic searches! When to use (ALWAYS try this first): - User wants papers on any topic: "find papers about brain-computer interfaces" - Simple keyword searches (you extract keywords from user query) - Basic year filtering if needed (one filter dimension) - Sorting by citations or date - ANY academic search unless it requires complex multi-dimensional filtering How to use: - query: Simple keywords (2-5 words) - YOU extract these from user's natural language Examples: "brain computer interface", "climate change mitigation", "CRISPR gene editing" - filters: OPTIONAL - Use sparingly, usually OMIT this Simple examples: {"from_publication_date": "2024-01-01"} or {"institutions.display_name": "MIT"} Database filter: {"locations.source.display_name": "PubMed"} or "arXiv" or "bioRxiv" - sort: "publication_date:desc" (newest first), "cited_by_count:desc" (most cited), or null for relevance - per_page: 25-200 (default 25) Returns: {meta: {count, per_page, page}, results: [{id, doi, title, authors, abstract, citations, pdf_url, institutions, credibility}]} 🎯 CRITICAL WORKFLOW: 1. User asks for papers → extract keywords → search_open_access_works (CHEAPEST) 2. If needs complex multi-filter → smart_open_access_search (EXPENSIVE) 3. Never use smart_open_access_search for simple queries Database-specific searches: Use filters to target specific databases: filters={"locations.source.display_name": "PubMed"/"arXiv"/"bioRxiv"} This replaces pubmed_search, arxiv_search, biorxiv_search for most cases! 💰 COST: Cheapest option - No extra API costs, no AI parsing needed.
smart_open_access_search
ChatGPT⚠️ EXPENSIVE TOOL - Use ONLY when complex multi-dimensional filtering is required. Prefer search_open_access_works for simple searches. What it does: - Uses AI (OpenAI API call - COSTS MONEY) to parse natural language into structured OpenAlex filters - Automatically determines complex filter combinations across multiple dimensions - Handles queries requiring simultaneous filtering by year + citations + institution + author + journal - Searches OpenAlex with AI-optimized parameters ⚠️ COST WARNING: Each call makes an OpenAI API request. Use sparingly! When to use (RARE cases only): - User needs COMPLEX filtering with 3+ dimensions that would be hard to manually specify - Query requires intelligent interpretation of ambiguous filter requirements - Example: "highly cited papers on AI from top universities published after 2020" (needs citation filter + institution filter + date filter) - User query has multiple contradictory or nuanced filtering requirements When NOT to use (MOST cases): - Simple keyword searches → use search_open_access_works (FREE, FAST) - Basic year filtering → use search_open_access_works with filters param - Single filter dimension → use search_open_access_works - User just wants "papers about X" → use search_open_access_works How to use: - query: Natural language research question (AI will parse it) - per_page/page: Pagination (defaults fine) Returns: Same as search_open_access_works - list of papers with metadata. 🎯 PRIORITY: ALWAYS try search_open_access_works FIRST. Only use this if the query genuinely requires complex multi-dimensional filtering that you cannot construct manually.
wisebase_file_list
ChatGPTget file list from Wisebase
wisebase_rag_answer
ChatGPTAn agentic RAG (Retrieval-Augmented Generation) system for iterative, multi-round knowledge retrieval. ⚠️ CRITICAL: This is NOT a single-call tool. You MUST call it multiple times (up to 20 rounds allowed). MANDATORY ITERATION PROTOCOL: 1. Call the tool with your initial query 2. Receive 'passages' and 'should_continue' flag 3. IF 'should_continue' is TRUE → YOU MUST CALL AGAIN WITH A DIFFERENT QUERY - Read and analyze ALL passages to understand what information you got - Identify what's MISSING or unclear from current passages - Generate a COMPLETELY DIFFERENT query from a NEW ANGLE - Call the tool again immediately - Do NOT answer the user's question yet 4. IF 'should_continue' is FALSE → You have enough information to answer 🔥 CRITICAL: Query Diversity Requirements Each new query MUST be SUBSTANTIALLY DIFFERENT from previous ones to avoid duplicate results: ❌ BAD (too similar): - Round 1: "user book acquisition" - Round 2: "user acquiring books" ← Same concept, will get duplicate results! ✅ GOOD (different angles): - Round 1: "user book acquisition" → general discussion - Round 2: "PDF reader features personalization" → technical features angle - Round 3: "reading behavior patterns analysis" → behavioral angle - Round 4: "book recommendation algorithms" → algorithmic angle How to generate DIVERSE queries: 1. Different aspects: If Round 1 was about "what", try "how", "why", "when", "who" 2. Different granularity: Broad overview → specific details → edge cases 3. Different perspectives: User view → system view → business view 4. Different domains: Features → behavior → architecture → performance 5. Complementary info: If got "benefits", ask about "challenges" or "implementation" Required analysis before each new query: Before calling again, you MUST: 1. Summarize key themes from current passages 2. List what topics/angles are MISSING 3. Choose ONE missing angle to explore 4. Formulate query targeting ONLY that new angle 5. Ensure new query uses DIFFERENT keywords than before Example multi-round with diverse queries: Round 1: "user book acquisition methods" → Got: General discussion about users getting books ✓ Covered: Basic concepts, user workflows ✗ Missing: Specific features, technical implementation Round 2: "PDF reading features personalization customization" → Got: Feature descriptions, customization options ✓ Covered: Features, UI/UX aspects ✗ Missing: User behavior data, usage patterns Round 3: "reading behavior PDF vs web statistics" → Got: Behavioral comparisons, usage statistics ✓ Covered: User behavior, comparison data ✗ Missing: Technical architecture, system design Round 4: "knowledge base architecture search retrieval" → Got: System design, technical implementation ✓ should_continue = false → NOW comprehensive coverage REMEMBER: - Same query = same results = wasted call! - Each query should explore a NEW dimension/angle - Analyze passages content, not just count - should_continue: true = MANDATORY different query - should_continue: false = Stop and answer - more then 5 rounds allowed
wisebase_upload
ChatGPTupload file to Wisebase
