Create the actual logger for operations that need immediate response

Ensure the required packages for a database type are installed. Args: db_type: The type of database backend

Helper to extract query texts from parsed JSON structures.

Fall back to the original LLM-as-query-planner approach.

Force flush all real loggers

Converts the Elasticsearch response in a list of values [url, schema_json, name, site_name] Args: response (List[Dict[st

Format MCP search results into the expected format for NLWeb. Args: mcp_result: Raw result from MCP search_shop_catalog

Return default configuration if file loading fails

Generate filename for saving HTML content.

Get item using URL-based retrieval.

Extract a unique identifier from an item.

Lazily load and return the provider for the given LLM type. Args: llm_type: The type of LLM provider to load Returns: T

Get or create the real LoggerUtility instance

Generate a unique hash for a URL.

Check if an endpoint has valid credentials based on its database type. Args: name: Endpoint name config: End

Load tools from XML file for a specific site. Args: tools_xml_path: Path to the tools.xml file site_id: The site ID to load tools

Try to find the full prefixed name for a property if given without prefix.

Column reference for GROUP BY / ORDER BY (no alias).

Fall back to LLM query planner

Fall back to LLM query planner

Search in Elasticsearch using kNN and filter Args: index_name: The index name embedding: The vector embeddin

Internal method to search with a single query. Args: query: The search query string site: Site identifier

Send error message when API call fails.

Send an error message to the client.

Send headers if they haven't been sent yet. Handles both streaming and non-streaming modes. Args: is_streaming:

Send message when no cricket data is found.

Send a single result to the browser.

Check all abort conditions and set the abort event if needed. Returns True if fast track was aborted, False otherwise.

Add a conversation to storage. Args: user_id: User ID (if logged in) or anonymous ID site: Site context for

Add a small response to the tracking list, maintaining a reasonable size

Analyze and print query results

Run query and analyze results including endpoint statistics

Make a request to an NLWebServer endpoint and parse the response. Args: endpoint: The full URL of the endpoint (e.g., "http://localh

Count model's top-k items where ref score < threshold.

Count how many of model's top-k have ref_score below threshold.

Categorize stores by their type.

Check Azure Embedding API connectivity

Check Azure OpenAI API connectivity

Check Azure AI Search connectivity

Check if messages were received in correct order.

Check OpenAI API connectivity

Clear all caches

Cleanup connections

Close the connection pool when done

Close the Elasticsearch client connections

Collate object attributes into a dict of lists.

Compute satisficer metrics for BERT model on a data split.

Compute metrics for a candidate model against a reference. If difficulty_filter is provided, only queries matching that category are included.

Compute rubric features that mirror GPT-4.1's relevance assessment. Returns a dict of float features, all cheap to compute from the schema JSON.

Ensure Gemini API is configured

Ensure Gemini API is configured.

Create and configure the aiohttp application

Create an assistant message with search results. Args: results: List of search results handler: Handler instance (provides c

Create a document in db_load.py compatible format. Args: store_data: Original store data embedding: Computed

Create a message from this participant.

Create valid email-based token.

Log a critical message asynchronously.

Ensure that a collection exists, creating it if necessary. Args: collection_name: Name of the collection to check/create

Ensure that the specified index exists, creating it if necessary. Args: index_name: Name of the index (defaults to confi

Log an error message asynchronously.

Log an exception with traceback asynchronously.

Extract [CLS] embeddings from trained BERT model.

Fetch content from a URL. Args: url: URL to fetch Returns: Tuple of (content, file_extension)

Clean up the response

* Formats an object as HTML * * @param {Object} obj - The object to format * @param {number} indent - The indentation level * @returns {s

Run full evaluation: metrics + per-site breakdown + satisficer.

Retrieve the API key for Gemini API.

Configure and return an asynchronous OpenAI client.

Get or initialize the Azure OpenAI client.

Generate embeddings for multiple texts using Elasticsearch Inference API. Args: texts: List of texts to embed

Get collection name from endpoint config or use default.

Retrieve all conversations in a thread.

Retrieve the Cortex Search Service (database, schema, service) to use from the configuration, or raise an error.

Generate an embedding for a single text using Elasticsearch Inference API. Args: text: The text to embed mod