Validate that all imports in code are in the whitelist. Args: code: Python source code to validate whitelist: List of al

Recursively add files and directories to the tree.

Answer *question* about *image_path*.

Answers a specific question about the content of the video file by analyzing keyframes and optionally its audio transcription, using multimodal G

Clean up resources and close tool server connections.

Connect to MCP tool servers specified by script paths. Args: scripts: List of paths to tool server scripts Rais

Execute a shell command in the workspace directory. Args: command: The shell command to execute.

Return a detailed caption of *image_path*.

r"""Check if the Docker daemon is running. Returns: bool: True if the Docker daemon is running, False otherwise.

Main async function that sets up and runs the AgentFly client. Args: args: Parsed command line arguments

Parse command line arguments for the AgentFly client. Returns: Parsed arguments namespace

Process a user query through the hierarchical AI system. This is the main method that: 1. Gets the meta-planner to break dow

r"""Executes a code file in a subprocess and captures its output. Args: file (Path): The path object of the file to run.

Trim message history to fit within token limit while preserving system message. Args: messages: List of message dictionaries

r"""Destructor for the E2BInterpreter class. This method ensures that the e2b sandbox is killed when the interpreter is deleted.

r"""Clean up the kernel and client.

r"""Destructor for the DockerInterpreter class. This method ensures that the Docker container is removed when the interpreter is dele

Initialize OpenAI backend with specified model. Args: model: The OpenAI model to use (e.g., 'gpt-4', 'o3')

Initialize the hierarchical client. Args: meta_model: Model name for the meta-planner agent exec_model: Mode

Return a + b.

Send chat completion request to OpenAI with optional tool calling. Args: messages: List of message dictionaries with rol

Crawl a URL to Markdown and extract only the content relevant to the query. Parameters ---------- url : str Target URL to cr

Deep crawl and extract key content from a web page (Markdown format). This tool is designed to perform *deep analysis* on a specific link ret

Return a / b, rounded to *decimal_places* (default 2).

Downloads a video from the given URL using yt_dlp and returns the local file path. Parameters: - url (str): The URL of the video to down

Execute Python code with import whitelist validation and save to specified filename. 🔧 EXECUTION PROCESS: 1. Val

Execute terminal command in workspace directory context. Workflow Note: To run a script (e.g., Python, shell), first create it using `wr

Return a Markdown‑rich description of *document_path* (xls/xlsx/csv).

Fetch cleaned Markdown content from a web page via Jina Reader API. This tool uses Jina Reader (r.jina.ai) to fetch and clean web page content,

Reads and returns the raw bytes of the video file located at *video_path*. Parameters: - video_path (str): The local path to the video f

Captures uniformly distributed screenshots from the video file and returns them as base64-encoded JPEG strings. Parameters: - video_path

Get comprehensive information about the workspace. 📊 INFORMATION PROVIDED: • Workspace directory path • Total files and

Get detailed file structure and directory tree of the workspace. 🎯 WHEN TO USE THIS FUNCTION: ✅ USE get_workspace_structure() WHEN:

Run a Google search via SerpAPI and return the organic results. Parameters ---------- query : str The search query strin

Generates a detailed and descriptive caption of the image located at *image_path*. Parameters: - image_path (str): The file path or URL

List all files in the unified workspace directory with tree structure using rich. Workflow Note: After listing files, you can read a fil