Record execution step

Search for relevant chunks using semantic similarity. Args: query: Search query top_k: Number of top

Add content to memory. Args: content: The content to add priority: Priority level of the content

Call the LLM and return parsed JSON response.

Run all enabled evaluators on a single result.

Helper method to log using task_log if available, otherwise skip logging.

Async version of cleanup_temp_file for use in async contexts. Args: file_path: Path to file is_temp: Whether file is tem

Save as a single JSON file

Send tool_call event

Send streaming update in new SSE protocol format

Connect to the MCP server and initialize the session.

Normalize title: fullwidth-to-halfwidth conversion, whitespace cleanup, lowercase.

Run the complete analysis and return summary statistics

Extract video metadata: duration, resolution, fps, etc.

Call a tool while maintaining the session.

Ensure file is available locally, downloading if it's a URL. Args: file_path: Local file path or URL Returns:

Analyze an image with optional multi-turn verification. Args: image_path_or_url: Local file path or URL of the i

Fuzzy title matching with normalization, keyword overlap, and edit distance.

Get UTC+8 timezone current time string

Extract audio metadata: duration, sample rate, etc.

Unified LLM call and logging processing Returns: Tuple[Optional[str], bool, Optional[Any], List[Dict[str, Any]]]:

Args: Returns:

Close client connection

Execute a single tool call. :param server_name: Server name :param tool_name: Tool name :param arguments: Tool argume

Check if compression is needed to fit incoming content

Check if the current embedding model is a Qwen/Alibaba model.

Send end_of_agent event

Send end_of_llm event

Send start_of_agent event

Send start_of_llm event

Main entry point for enhanced video understanding. Args: video_path: Path to video file (local or URL) quest

Create a visual progress bar for percentage display

Find the earliest start time from all completed files

Find the latest end time from all completed files

Get or create singleton audio client

Get or create singleton video client

Get or create singleton VQA client

Normalize a string by: - Removing all white spaces - Optionally removing punctuation (if remove_punct is True) - Conv

Build the prompt for VLLM.

Encode audio to base64 and determine format

Extracts the query string from tool call arguments based on tool_name. Supports search_and_browse, google_search, sougou_search, and

Get or create RAG instance for the given JSON path.

Initialize SQLite database for embedding cache. Args: db_path: Path to the SQLite database file read

Load chunks and embeddings from cache.

Parse VLLM response and extract structured information.

Attempt to fix truncated JSON by closing unclosed brackets/braces.

Verify claims against video content using Gemini multimodal model.