Extract [CLS] embeddings without the regression head.

Generate the path for the equivalent file with embeddings. Args: file_path: Path to the original file Returns: Path to

Convert a filter spec to SQL WHERE clause fragment.

Extract name from a JSON item using various fields. Args: item: JSON item to extract name from Returns: Name string or

Get the specified LLM provider config or the preferred one if not specified.

Create and return a configured logger for the specified module

Get memory usage statistics.

Get or initialize the Ollama client.

Extract podcast details including RSS feed and organization

Get settings for a specific profile

Get or initialize Qdrant client.

Convert values to ranks (1-based, higher value = lower rank number).

Convert values to ranks (1-based, higher value = lower rank number).

Retrieve the N most recent conversations for a user and site, grouped by thread. Args: user_id: User ID to retrieve conv

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

Handle an MCP request following the standard MCP protocol Args: query_params (dict): URL query parameters body (bytes): Requ

* Unescapes HTML entities in a string, safely converting entities like & to & * without executing any HTML/scripts. * * @param {string}

Check if the specified index exists. Args: index_name: Name of the index to check (defaults to configured index)

Initialize backends

Compute a hash value for a string, ensuring it fits within int64 range. Args: string: The string to hash Returns: int64

Check if aggregation functionality is enabled.

Check if score_openrouter.py is currently running.

Returns True if the system is running in testing mode.

Check if the who endpoint is enabled.

Join an existing conversation.

Leave the conversation.

Load data from a file, compute embeddings, and store in the database. The file can have one of multiple formats: 1. Two columns per row,

Load conversation storage configuration.

Log a message with additional context information asynchronously.

Map variable and place names to Data Commons DCIDs.

Parse XML sitemap to extract URLs.

Compute Spearman ρ per group.

Plot results and save to file.

Prepare documents from URL and JSON data. Args: url: URL for the item json_data: JSON data for the item site: Site i

Prepare the streaming response

Print summary of aggregated test results.

Print detailed results for a single test.

Print the current status

Print summary of all query results

* Handle message processing for different message types

Process all sites in the JSONL file.

Process a file in binary mode for cases where text encoding is problematic. Specifically handles UTF-16 encoded files.

Process a cricket statistics JSON response and return formatted table. Args: json_response: JSON string from cricket statistics API

Main method to process cricket JSON response and return formatted output.

Process an RSS/Atom feed into document objects. Args: file_path: Path to the RSS file or URL site: Site identifier Retu

Read a YAML configuration file.

Receive a message.

Reconnect with optional sequence ID for sync.

Run a set of queries and return all results Args: queries: List of dicts with 'query' and optional 'site', 'num_results'

Run standard query tests with endpoint tracking

Run standard query tests

Run tests from a JSON file. Args: file_path: Path to JSON test file test_type: Optional test type override (

* Sanitizes a URL to prevent javascript: protocol and other potentially dangerous URLs * * @param {string} url - The URL to sanitize * @retu

Delegate to shared loss_utils.satisficer_metrics (includes pairwise accuracy). Also computes top_k_overlap and high_bad for full evaluation.

Compute satisficer-style metrics: top-k overlap, good/bad picks.

Save enhanced sites to JSONL file.

Save current progress to disk.

Save current state to disk.

Search for sites. Returns: List of {"url": str, "json_ld": str, "name": str, "site": str}

Search for documents matching the query and site using vector similarity. Args: query: Search query string s

Retrieve a document by its exact URL. Args: url: URL to search for endpoint_name: Optional name of the endpoint to use

Retrieve a document by its exact URL. Args: url: URL to search for endpoint_name: Optional endpoint name ove

Send ranked sites to the client.

Send begin/start message at the start of query processing.

Send error response - for compatibility

Set all loggers to the same level

Synchronous version for compatibility

Set read operation latency.

Simulate network latency.

Start the server in a background thread

Stop the server gracefully

Stop the server

Remove site name references, URLs, and publisher info for site-agnostic training.

Convert test case to dictionary.

Count overlap of top-k items.

Count how many of the top-k items by ref_scores are in the top-k by model_scores.

Trim schema.org JSON to remove unnecessary fields. This matches NLWeb's trim_json function exactly.