Stop watcher for a single project.

Upsert a single batch with retry logic.

Verify that storage count matches expectations.

Add a file change.

Add a project to the watch list.

Analyze chunk type distribution and storage efficiency

Assess content quality - identify generic/duplicate names and minimal information value

Compare entity counts across categories, ratios, and relation density

Identify files generating excessive entities (>50-100 entities)

Extract manual entries from any Qdrant collection and save to backup file.

Build the prompt for comprehensive code quality analysis.

Call Claude CLI for comprehensive code quality analysis.

Check if this parser can handle the given file.

Check if code contains override comments to allow duplicates.

Clean up relations orphaned by content hash changes Note: For 66x performance improvement, use: - self._get_qdrant_store()._get_all_r

Remove old entries to prevent memory leaks.

Clear the log file for a collection.

Compare the two file sets and return differences.

Create BM25 embedder with default configuration.

Create a vector point from a ChatChunk for v2.4 pure architecture.

Create a new Qdrant collection with both dense and sparse vector support. Args: collection_name: Name of the collection t

Create an embedder instance from configuration.

Create a hybrid vector point from an EntityChunk with both dense and sparse embeddings. Args: chunk: EntityChunk to creat

Create metadata chunk from existing Entity for progressive disclosure.

Create a vector point from a RelationChunk for v2.4 pure architecture.

Create a vector store instance from configuration.

Decode project directory name back to original path.

Create empty context for cases where Git+Meta is not applicable.

Encode project path to match Claude's directory naming.

Extract user and assistant text messages from Claude Code JSONL chat file

Extract observations for class entities.

Extract actual code content from formatted code_info. Args: code_info: Formatted code info string (may contain ``` block

Extract entity names from Write/Edit operations.

Split relations into changed vs unchanged based on entity involvement

Find ALL manual entries (excluding documentation) to format with [ ].

Find all entities associated with a file path using OR logic. Searches for: - Entities with file_path matching the given path

Find specific implementations missing metadata chunks

Force return all pending files for cleanup.

Create config with environment variable overrides.

Get list of all collections.

Get all project directories from Claude projects directory.

Extract code information from the operation.

Get collection statistics using direct Qdrant API methods.

Get current global Memory Guard status.

Get detailed status of git hooks.

Get information about the embedding model.

Map project path to Claude chat directory.

Get all file entities from Qdrant collection.

Get all tracked files from indexer state files.

Get service status.

Get list of supported file extensions.

Get supported extensions.

Get supported extensions.

Generate a unique test collection name with timestamp.

Check if content contains function or class definitions. Args: content: Code content string Returns:

Check if there are pending files.

Index a single file with Git+Meta deduplication.

Install pre-commit hook for automatic indexing.

Check if lines contain only configuration constants. Args: lines: List of code lines Returns: True

Check if the pre-commit hook is installed.

Check if all non-empty lines are import statements. Args: lines: List of code lines Returns: True i

Check if conversation has been inactive for threshold hours.

Enhanced logic for v2.4 chunk format - matches manual_memory_backup.py exactly.

Check if lines contain only simple variable assignments. Args: lines: List of code lines Returns: T

Check if operation is trivial and should skip guard analysis.

Enhanced logic for v2.4 chunk format. Uses the same detection logic as qdrant_stats.py for consistency.

Load session IDs of existing chats from chats.md.

Load settings from the project's settings.txt file.

Load URL content hashes from existing stats file

Main entry point using Click CLI.

Main entry point for the hook.

Main function to find missing files.

Main execution function

Main function.

Main entry point.

Main CLI interface.

Convert absolute path to relative path from project root.

Extract HTML structure, IDs, classes, components.

Parse Claude's response to extract duplicate detection result.

Parse content with appropriate language based on file extension.

Print detailed comparison results.

Single entry point replacing _store_vectors() logic.

Process the hook and return the result.

Register a storage backend.

Remove navigation and UI noise from markdown content

Directly restore manual entries to Qdrant with proper vectorization. This function bypasses MCP and directly inserts entities into Qdrant with em

Run indexing with the specified parameters. This function provides a programmatic interface for other modules. It discovers files and delegat

Run deletion handling with shared deletion logic for a single file.

Save URL content hashes immediately with file locking

Save scraped content to markdown files

Save single MD file immediately after processing

Scrape all sections defined in the site config

Search for similar vectors.

Determine if this hook event should be processed.

Check if a file should be processed based on patterns and constraints. Args: file_path: Path to the file to check project_path: R