Analyze media using AI model

Build prompt messages for analysis

Direct call to builtins like open, eval, exec with no import.

Check for required dependencies

Compose final generation output.

将字符串值转换为适当的类型

Download remote media file with validation

Extract audio from video

Extract frames based on scene changes

Extract key frames intelligently

Get file extension from content type or URL

Handle calls of the form object.method(...) or module_name.function_name(...).

If the function is simply called by name (e.g. open, remove, etc.), figure out which module it's from (if any) and check if i

Initialize configuration with sensible defaults

Initialize HTTP client with retry logic

Detect if file is audio-only

Parse and validate parameters

Process input URL/path and get local file path

Process next observations from environment.

Resolve local file path, handling relative paths

保存缓存到文件

保存缓存到文件

Stream file content to disk with progress tracking

Transcribe audio to text

Update rolling state with new responses and observations.

Validate media file existence and size

Aggregate results from multiple rounds

Orchestrates visiting and processing web pages. If a single URL is provided, it's processed directly. If a list of URLs is pr