End-to-end test participant (human or AI).

WebSocket client for E2E testing.

Core message structure for all communication. Separates WHO sent it (sender_type) from WHAT it contains (message_type).

Represents a single conversation entry (one exchange between user and assistant). Used for storing conversation history in the database.

Small MLP for a single feature's shape function f_i(x_i).

Mock WebSocket client for authentication testing.

Dataset for ModernBERT scorer training with rubric features.

Unified client for vector database operations. This class routes operations to the appropriate client implementation based on the database ty

ModernBERT with optional rubric features and a regression head. Architecture (use_rubric_features=True): ModernBERT encoder → [CLS] (hidd

Dataset for RubricGAM training. Expects pre-computed BERT embeddings + rubric features grouped by category.

Main aiohttp server implementation for NLWeb

Simulated client for load testing.

A configurable logging utility with different verbosity levels.

Executes template-based query plans.

Qdrant-based storage for conversation history.

Rubric-based Neural GAM with BERT embeddings. score = sigmoid(bias + f_bert(proj(cls)) + Σ f_group(rubric_features)) Supports additive, gate

Dataset for Neural GAM training. Expects pre-computed BERT embeddings + handcrafted features.

Real WebSocket client for testing.

Enumeration of test types.

Wrapper to make aiohttp StreamResponse compatible with existing NLWeb handlers. Provides the same interface as the original HandleRequest cla

Raised when configuration is missing or invalid

Convert cricket statistics JSON responses to tabular format for LLM processing.

Client for Elasticsearch operations, providing a unified interface for vector-based search results using the Elasticsearch Python client.

Elasticsearch-based storage for conversation history.

Test result for end-to-end tests.

Test runner for end-to-end query tests.

Extended VectorDBClient that tracks endpoint statistics

Client for PostgreSQL vector database operations with pgvector extension. Provides a unified interface for indexing, storing, and retrieving

Class to run prompts with a given handler.

Test result for query retrieval tests.

Test runner for query retrieval tests.

Store information about problematic responses for analysis

Client for Shopify MCP operations, providing search functionality via MCP endpoints. Currently supports search_shop_catalog method with USD c

Test result for site retrieval tests.

Test runner for site retrieval tests.

Simple TTL cache with LRU eviction

Base test case data structure.

Base test result data structure.

User query content structure.

Represents a single (query, item, model) scoring task.

Basic handler for A2A protocol requests

Anthropic Claude implementation (placeholder)

Implementation of LLMProvider for Anthropic API.

HTTP server that proxies NLWeb `/ask` and wraps responses for AppSDK.

Background processor for handling log writes asynchronously

Azure OpenAI implementation

Implementation of LLMProvider for Azure OpenAI.

Azure AI Search implementation

Client for Azure AI Search operations, providing a unified interface for indexing, storing, and retrieving vector-based search results.

Azure AI Search-based storage for conversation history.

Client for Bing Web Search API operations, providing read-only access to web search results with optional site-specific filtering.

Raised when configuration is missing or invalid.

Raised when configuration is missing or invalid.

Raised when configuration is missing or invalid.

Custom ranking for conversations that includes user_id in results.

Implementation of LLMProvider for DeepSeek on Azure.

Manage end-to-end conversation testing.

Elasticsearch implementation (placeholder for future implementation)

Test case for end-to-end tests.

Implementation of LLMProvider for Google's Gemini API.

Implementation of LLMProvider for Hugging Face Inference Providers.

Implementation of LLMProvider for Inception API. Perform a single-shot (non-streaming) chat completion asynchronously. Returns the full a

Incremental crawler that maintains state and processes URLs one by one.

Main class for enhancing JSONL site descriptions.

Handles LLM API calls for description enhancement.

Lazy logger that defers actual logger creation until first use and writes asynchronously

Implementation of LLMProvider for Llama on Azure.

Monitor system resources and performance during load tests.

Helper class to load and manage logging configuration from YAML file

Handler for standard MCP protocol requests

Loads JSON-LD mapping files and provides lookup methods for the SQL compiler.

Helper class for sending messages in NLWebHandler. This class encapsulates message sending utilities to reduce clutter in the main NLWeb

Track message delivery and ordering.

Type/purpose of the message content.