Set browser variables to corresponding link, page HTML for URL

Runs the QLLM agent synchronously in a new thread. NOTE: This sync interface is solely because the CaMeL interpreter is synchronous a

Wrapper for `receive` handler for performing search action on current page

Save the current session to disk. Returns the file path, or None if nothing to save.

Detect content type and apply appropriate formatting.

Core operator dispatch.

Mock cleanup.

Save artifact to output folder

Interprets the given AST enforcing security policies.

Create standard configuration for Gemini content generation. Args: temperature: Temperature for generation (default: 1) top_p: To

Run the tool asynchronously.

Load JSON file, return empty dict if not found.

Submits the current solution and ends the interaction. Don\'t submit unless you are confident that the solution is correct. For examp

Mock command execution.

Save an image to the debug folder with a descriptive name. Args: img_bytes: Image as bytes step_name: Name of the processing step

Returns a standard Python logger.

Standard Reasoning Engine entry point.

Generate content using Gemini image generation model. Args: client: Gemini client instance text_images_pieces: List of text strin

Evaluate all rules against the provisional output using the Collect -> Plan -> Execute pipeline. This is the main entry

Resolve a dot-notation field path against a nested dict. Examples: resolve_field({"a": {"b": 1}}, "a.b") → 1

Runs the agent to solve the benchmark task. Args: max_turns: The maximum number of turns to allow the agent to run. m

Get a config value from environment, raising if missing and no default.

Generate text content using Gemini. Args: text_images_pieces: List of text strings, image bytes, or GCS paths client: Gemini clie

Create an MP4 video in memory and return its raw bytes. Args: frames_bytes: List of frame bytes fps: Frames per second (default:

Execute a single action on the output. For llm_continue_processing, this delegates to LLMActionExecutor which calls Gemi

Detect MIME type from file extension in path. Args: path: File path or URL Returns: str: Detected MIME type

Loads an image from various sources and inserts it into a slide. If a box_hint is provided, it scales the image to fit within the box whi

Execute a function in parallel across multiple inputs using ThreadPoolExecutor. Args: to_predict: Iterable of inputs to process p

Classify shoe position using remote endpoint or local embeddings fallback.

Convert input to appropriate Part type (text, image/video bytes, or GCS path). Auto-detects MIME type from bytes or file extension. Args:

Load case data and build context. Args: case_id: The case folder name (numeric ID). Returns:

Begin a new session.

Recursive markdown rendering. depth controls heading level for nested structures (so a top-level object's nested keys get h2, the next level h

Save data to JSON file.

Upscales an image using GCP's Imagen 4.0 Upscale API. Automatically adjusts upscale factor if the requested factor would exceed the maximum a

Validate a rule dict against the schema. Returns list of error messages (empty if valid).

Wrapper for `receive` handler for performing click action on current page

Detect MIME type from file signature (magic bytes). Args: data: Bytes to analyze Returns: str: Detected MIME type (e.g., "im

Strip non-content elements from raw HTML.

Return the module-level master data instance (may be None).

Set a field value using dot-notation path.

Write rule base to temp file and return path.

Create a canvas with a product image centered with configurable margins. Supports three scaling modes: - target_diagonal: Scales image to mat

Retrieves the value of an environment variable. Args: var_name: The name of the environment variable. Returns: The value of the

Normalize a string for comparison (lowercase, strip, collapse whitespace).

Extract the product object from the image and place it on a new background. Handles EXIF orientation and ensures mask alignment with the original

Return the current CONFIG dict.

True if a value carries no information worth rendering.

Generates detailed content for MULTIPLE slides in parallel with performance tuning.

Compare two face images using InsightFace embeddings. Args: reference_face_bytes: Reference face image as bytes. generated_face_b

Extract frames from video bytes and return them as a list of image bytes. Args: video_bytes: The raw byte data of a video file im

Generate video using Veo with reference images (R2V mode). Args: client: Veo client instance reference_images: List of reference

Create standard configuration for Gemini content generation.

Loads nurse profiles from the mock HRIS database (internal use).

Detect a face in the image using Google Cloud Vision API and return the cropped face. Padding is calculated as a percentage of the detected face

Get a logger instance with proper configuration.

Converts the Customer object to a JSON string. Returns: A JSON string representing the Customer object.

Clean LLM JSON response by removing markdown code blocks.

Execute a command in the container with timeout and error handling. Args: command: The command to execute. demux: If True, re

* Helper function to read environment variables. * @param key The environment variable key. * @param defaultValue The default value if the key is no

Detect MIME type from file signature (magic bytes). Needs at least 32 bytes to check every signature reliably; shorter inputs are treated as

Initializes or retrieves the YouTube Data API client. Priority: 1. Session state (onboarded by user) 2. Environment variable (pr

Parse a document file to markdown. Args: file_path: Local path to the file. genai_client: Gemini client (needed for PDF/legacy Of

Upload an entire folder to GCS using the High-Performance Transfer Manager. Args: bucket_name: Name of the GCS bucket source_fold

Check if the labels contain enough views to generate a video. Args: labels: List of class labels. Returns: bool: True if eno

Wrap agent output with metadata

Generates mock shifts for the scheduling period. Args: start_date: Starting date for shifts (defaults to today) num_days: Nu

Generate video using Veo from a starting image. Supports two modes: - Image-to-video: Only provide `image`, Veo animates from that frame

Detect MIME type from file extension.

Logs the execution of a policy to Firestore and updates aggregate stats. Args: violations: A list of violation objects. If provided,