Skip to main content

Tool Calling

A Tool Calling is a function that the LLM can invoke autonomously during a conversation to retrieve information or perform an action. Instead of relying purely on training data, the LLM calls tools to fetch live data, search indexed content, execute code, and more — then incorporates the results into its reasoning.

Turing ES includes 27 native tools organized into 7 categories, plus support for external tools via MCP Servers. Tools are enabled per AI Agent — each agent selects only the tools it needs.

Semantic Navigation — 15 tools​

These tools allow the LLM to interact with any Semantic Navigation Site as a structured knowledge source, enabling rich search-based reasoning over your indexed content.

ToolDescription
list_sitesLists all available SN Sites with their locales
get_site_fieldsReturns valid facet fields for filtering a specific site
get_valid_filter_valuesReturns valid values for a specific filter or facet
search_siteSearches a site and returns compact results (ID, title, URL, snippet)
get_document_detailsRetrieves full text and metadata of a document by ID
get_search_suggestionsAutocomplete and spelling corrections for a search query
find_similar_documentsFinds semantically similar documents (More Like This)
get_aggregated_statsCalculates totals and distributions by category via facet aggregation
get_document_highlightsExtracts snippets from a document where search terms appear
compare_itemsCompares specific fields of two or more documents side by side
search_recent_updatesRetrieves the most recently updated content on a topic
get_facet_summaryStatistical summary of all available categories and attributes
search_by_date_rangeSearches documents within a date range
lookup_facet_valueSearches a term across all facets to find exact values
discover_facet_valuesSplits a phrase into words and searches across all facets

Prompt examples:

  • "Search for documents about authentication in the Sample site" → triggers search_site
  • "What fields can I filter by on the Sample site?" → triggers get_site_fields, then get_valid_filter_values
  • "Show me the full content of document ID abc-123" → triggers get_document_details
  • "Find documents similar to the article about Solr configuration" → triggers find_similar_documents
  • "How many documents do we have per content type?" → triggers get_aggregated_stats or get_facet_summary
  • "What articles were updated in the last 7 days about security?" → triggers search_recent_updates or search_by_date_range
  • "Compare the features of Product A and Product B" → triggers search_site then compare_items

RAG / Knowledge Base — 4 tools​

These tools enable the LLM to query the Knowledge Base — files indexed from MinIO via the Assets section — using semantic similarity search.

ToolDescription
search_knowledge_baseSearches for relevant documents by semantic similarity (top 10, threshold 0.7)
knowledge_base_statsReturns statistics: total files, chunks, and storage size
list_knowledge_base_filesLists all indexed files, with optional keyword filter
get_file_from_knowledge_baseRetrieves the full indexed content of a specific file

Prompt examples:

  • "What does our internal documentation say about deployment procedures?" → triggers search_knowledge_base
  • "How many files are indexed in the knowledge base?" → triggers knowledge_base_stats
  • "Show me the full content of the file onboarding-guide.pdf" → triggers list_knowledge_base_files then get_file_from_knowledge_base

Web Crawler — 2 tools​

ToolDescription
fetch_webpageFetches a web page by URL and returns its content as plain text
extract_linksExtracts all links from a web page, with optional keyword filter

Prompt examples:

  • "What does the Apache Solr documentation say about faceted search?" → triggers fetch_webpage
  • "List all links on our company blog homepage" → triggers extract_links

Finance — 2 tools​

ToolDescription
get_stock_quoteCurrent price and market data for a stock ticker symbol
search_tickerLooks up a ticker symbol by company name or keyword

Prompt examples:

  • "What is the current stock price of Apple?" → triggers search_ticker then get_stock_quote
  • "Show me the market data for MSFT" → triggers get_stock_quote

Weather — 1 tool​

ToolDescription
get_weatherCurrent weather and forecast for a city (1–7 day range)

Prompt examples:

  • "What's the weather forecast for São Paulo this week?" → triggers get_weather

Image Search — 1 tool​

ToolDescription
search_imagesSearches the web for images and returns URLs and descriptions

Prompt examples:

  • "Find images of enterprise search architecture diagrams" → triggers search_images

DateTime — 1 tool​

ToolDescription
get_current_timeReturns the current date and time for a given IANA timezone

Prompt examples:

  • "What time is it right now in Tokyo?" → triggers get_current_time

Code Interpreter — 1 tool​

ToolDescription
execute_pythonExecutes Python code in a sandboxed environment and returns stdout/stderr

The Code Interpreter runs Python in an isolated sandbox directory with a 30-second execution timeout. It supports:

  • Standard output capture
  • Matplotlib chart generation (with Agg backend — no display required)
  • Automatic print() wrapping for bare expressions

The Python executable path is configured in Administration → Settings → Python Path.

Prompt examples:

  • "Calculate the compound interest on $10,000 at 5% for 10 years" → triggers execute_python
  • "Generate a bar chart showing monthly sales: Jan=100, Feb=150, Mar=120" → triggers execute_python with matplotlib

External Tools via MCP Servers​

Beyond the 27 native tools, AI Agents can access tools from any external server implementing the Model Context Protocol (MCP). This covers company-internal systems, proprietary data APIs, and the growing ecosystem of public MCP servers.

See MCP Servers for configuration details.

PageDescription
AI AgentsHow to compose agents with the tools they need
MCP ServersExtend agents with external tools via MCP
AssetsKnowledge Base files queried by RAG tools
Semantic NavigationThe search experience powering SN tools