Unlock AI Data Lake Power with LAS CLI: 9 Operators in One Command

Overview

LAS CLI is a command‑line client for Volcano Engine's AI Data Lake (LAS) service. It provides a uniform las-cli operators call interface to invoke nine online operators that perform speech recognition, audio conversion, video editing, video understanding, image resampling, PDF parsing and other multimodal data‑processing tasks.

Key Features

Agent‑friendly syntax : a single command such as

las-cli operators call las_asr_pro --input_url https://example.com/audio.wav

replaces custom HTTP request code.

Low token consumption : the model only needs to fill parameter values, reducing LLM token usage.

Structured I/O : arguments follow a --param value pattern, eliminating manual JSON handling.

Pipeline composability : CLI output can be piped to subsequent commands, enabling chained workflows.

Protocol‑agnostic : the CLI can be wrapped by any higher‑level protocol or orchestration layer.

Minimal learning curve : all operators share the same call syntax.

Supported Operators

Audio

ASR Pro ( las_asr_pro): noise reduction, speaker separation, emotion detection, multilingual recognition.

Audio Convert ( las_audio_convert): lossless conversion among WAV, MP3, FLAC, etc.

Audio Extract & Split ( las_audio_extract_and_split): extract audio from video and split by duration.

Video

Video Edit ( las_video_edit): natural‑language driven video editing.

Video Inpaint ( las_video_inpaint): remove watermarks or subtitles with pixel‑replace or generation modes.

Video Resize ( las_video_resize): intelligent resolution adjustment with aspect‑ratio control.

VLM Video ( las_vlm_video): visual‑language model for deep video understanding.

Image

Image Resample ( las_image_resample): resampling with Nearest, Bilinear, Bicubic, Lanczos interpolations.

Document

PDF Parse Doubao ( las_pdf_parse_doubao): structured extraction from complex PDFs using the Doubao model.

Quick Start

Installation

If Node.js ≥ 16 and npm are available: npm install -g @las-ai/cli@latest Otherwise install the pre‑compiled binary:

curl -fsSL https://lf3-static.bytednsdoc.com/obj/eden-cn/0eh7lpqbog/foeh7bfbvo/install-cli.sh | sudo bash

Verify with las-cli --version.

Configuration

Run the interactive initializer and provide:

LAS API Key (from the LAS console).

Default region (e.g., cn-beijing or cn-shanghai).

Optional large‑model service endpoint, model name, and model API key for task create commands.

Optional TOS bucket address for storing processed files.

The settings are saved to ~/.las-cli/config.json.

Calling an Operator

Example – describe a video using the VLM Video operator:

las-cli operators call las_vlm_video \
  --video_url "https://p9-arcosite.byteimg.com/obj/tos-cn-i-goo7wpa0wc/e8d1effd969c4f67942b21cb8aa3bc19" \
  --text "Please describe the video content" \
  --temperature 0.7

The command returns a JSON object with the analysis result.

Advanced Usage

Task Orchestration

LAS CLI includes a built‑in task manager. After configuring a large‑model service, a natural‑language description can be submitted as a task:

las-cli task create "Analyze the video at tos://bucket/video.mp4, describe its content and extract the audio"

The CLI parses the description, selects the appropriate operators (e.g., las_vlm_video and las_audio_extract_and_split), executes them in order, and returns the final result. Task status can be queried with las-cli task list and las-cli task status <task-id>.

Pipeline Composition

Operators can be chained using Unix pipes or shell loops. Example batch workflow:

for f in ./videos/*.mp4; do \
  las-cli operators call las_vlm_video --video_url "$f" --text "Summarize" --out "${f%.mp4}_summary.json"; \
done

Output files can be stored in a TOS bucket by adding --output_bucket tos://my-bucket/ (if supported by the operator).

Automation & Testing

Regression testing:

las-cli operators call las_asr_pro --input_url https://example.com/audio.wav --out result.json

and compare the JSON against expected values.

Scheduled jobs: embed CLI calls in cron or workflow engines for daily audio transcription or weekly PDF parsing.

Monitoring: use the CLI exit code (0 = success) and parse the JSON output to generate health checks.