English

← Back to list

One line of commands to overturn 5 APIs! Alibaba Cloud Bailian CLI is open source, and the battle for water, electricity and coal between Agents has officially begun

itadol5j

cover

On May 29, Alibaba Cloud did something that brought a collective climax to Agent developers - Bailian's core capabilities were officially CLI-ified. One line of commands can allow AI Agents to automatically access more than 150 models, more than 10 applications, knowledge bases, memories, Internet search and other full capabilities.

More importantly, Bailian CLI has been open sourced on GitHub (modelstudioai/cli) and natively supports mainstream Agent frameworks such as Claude Code, Qoder, OpenClaw, and Hermes Agent.

To sum up in one sentence: In the past, you had to manually adjust 10 APIs to achieve the Agent capabilities, but now you can do it all with one line of bl command.

What exactly is Bailian CLI?

Alibaba Cloud Model Studio is a one-stop large model and Agent application development platform. In the past, its interaction method was Web console + API SDK, and you had to write code, configure the environment, and manage Token.

Now Bailian CLI has moved all core capabilities to the command line:

  • Text dialogue: Qwen3.7-max and other mainstream models
  • Multi-modal (Omni): text + image + audio + video fully supported
  • Image generation/editing: Qwen-Image 2.0, professional text rendering, photo-quality, multi-image synthesis
  • Video generation/editing: HappyHorse-1.0 series, Vincent video/Picture video/Reference video, up to 9 pictures for reference
  • Speech synthesis/recognition: CosyVoice streaming TTS + 5-20 second speech cloning; FunAudio-ASR covers 30 languages including 7 Chinese dialects
  • Image/video understanding: Qwen-VL, long video analysis, chart document analysis, visual reasoning, multi-language OCR
  • Knowledge base + memory: multi-modal RAG retrieval + cross-session memory
  • Application call: Directly call the Agent and Workflow published by Bailian platform
  • Internet Search: Real-time Internet Search
  • Console capabilities: browse application list, check free quota
  • Automatic upload of local files: Each URL parameter accepts a local path and is automatically uploaded to free temporary storage (valid for 48 hours)

And each command can be used as a structured tool call - this means that the AI Agent can be called directly without you writing a wrapper.

Three-step installation, take off immediately

Environment requirements: Node.js >= 22.12.

Step 1: Install CLI + Skills

npm install -g bailian-cli
npx skills add modelstudioai/skills --all -g

Step 2: Certification

bl auth login --api-key sk-xxxxx

The API Key is obtained from the Bailian console. Free users will receive 1 million Tokens for each model, up to a maximum of 50 million.

Step 3: Activate

Chat with Qwen:

bl text chat --message "What is DashScope?"

Multimodal dialogue (text + picture + audio + video):

bl omni --message "描述这张图" --image ./photo.jpg

Generate pictures:

bl image generate --prompt "A cat in a spacesuit" --out-dir ./images/

Generate video (from local image):

bl video generate --image ./cat.png --prompt "Make the cat move" --download cat.mp4

Check the free quota:

bl usage free --model qwen3-max

Browse Bailian apps:

bl app list

Since updated:

bl update

Generate a 2-minute Japanese youth video in one sentence: official real demo

illustration

There is an official demo on the GitHub README of Bailian CLI. I was confused after reading the entire demo.

Original prompt:

Help me generate a Japanese film and television style, a high school girl's first love story, the plot is sweet, it makes people want to fall in love after watching it, the video is about 2 minutes, the size is 16:9

The final output: a full 2-minute, 16:9 cinematic short film with zero manual editing.

The execution chain behind it is as follows:

  • Qwen Code (agentic coding model) analyzes user intentions, plans narrative rhythm, and decides which tools to call
  • spark-video Skill Split the story into shots, write prompts for each shot, and maintain visual coherence (characters, lighting, tone, shot language)
  • Bailian CLI's bl video generate distributes each shot in parallel to HappyHorse 1.0 generation
  • Skill re-spliced all the clips into the final 16:9, 2-minute finished product

There is no timeline dragging, no frame-by-frame editing, one sentence comes in and one video comes out.

This is the greatest value of CLI - turning complex workflows into atomic capabilities that can be automatically orchestrated by Agents.

Why CLI is the killer move in the Agent era

You may think: CLI is just the command line, it’s an old technology, why is it so exciting?

The key lies in the compatibility between the CLI format and the Agent.

Traditional API access:

  • You need to write code (one for each of Python/Node/Go)
  • Control the SDK version
  • To handle authentication, timeout, retry, error code
  • When calling Agent, you also need to write a wrapper

CLI access:

  • Agent uses shell to work, CLI is natural language
  • One line of command is a tool call, and the structured output is directly fed back to the model.
  • No pressure across languages, bash, Python, Node, Go can all be adjusted
  • Upgrading only requires bl update, and the Agent does not need to change the code.

What’s even more exciting is that Bailian CLI is directly and natively compatible with mainstream Agent frameworks such as Claude Code, Qoder, OpenClaw, and Hermes Agent. This means:

  • You said in Claude Code to help me generate an image, it directly calls bl image generate
  • You said to analyze this video in OpenClaw, it directly adjusts bl's video understanding ability
  • You are writing code in Qoder. If you lack the ability, just bl it.

Agent does not require any additional configuration to have the full set of capabilities of Bailian’s 150+ models.

What's behind 150+ models

The Bailian platform aggregates not only the Ali Tongyi series, but also includes:

  • Tongyi Qianwen full series (Qwen3-Max, Qwen-Image 2.0, Qwen-VL, Qwen Code, etc.)
  • DeepSeek Series
  • GLM series (wisdom spectrum)
  • Kimi Series (Dark Side of the Moon)
  • MiniMax Series
  • Alibaba self-developed multi-modal modes: HappyHorse, CosyVoice, FunAudio-ASR
  • Extensive integration of third-party open source models

The advantage of this aggregation is that you don’t need to apply for an API Key separately, recharge separately, or connect separately for each model company. One Bailian account takes all.

Combined with Alibaba Cloud's Coding Plan (fixed monthly subscription for multiple models in one), developers can use all head models at a fixed monthly cost and no longer have to worry about API calls exceeding the budget.

The authentication method is very considerate

illustration

Bailian CLI provides three authentication methods:

  • Environment variable: export DASHSCOPE_API_KEY=sk-xxxxx
  • Login command: bl auth login –api-key sk-xxxxx (persisted to ~/.bailian/config.json)
  • Single command: bl text chat –api-key sk-xxxxx –message Hello

Console capabilities (app list, usage free) require browser login:

bl auth login --console

Knowledge base retrieval requires RAM sub-account AccessKey:

export ALIBABA_CLOUD_ACCESS_KEY_ID=LTAI5t...
export ALIBABA_CLOUD_ACCESS_KEY_SECRET=...
export BAILIAN_WORKSPACE_ID=ws-...

The official also reminds: use RAM sub-accounts, do not use root accounts AK/SK - this security reminder is very professional.

The real signal of Ali’s layout

If you put Bailian CLI into Alibaba Cloud's overall actions this year, the rhythm will be very clear:

  • January 28: Alibaba Cloud launches a full set of OpenClaw cloud services, fully providing the infrastructure required by Agents
  • April: Tongyi Qianwen Qwen3.7-Max is released, and the programming ability is catching up with Claude
  • May: Coding Plan subscription package is launched, starting from 49 yuan/month to aggregate 4 major head models
  • May 29: Bailian CLI is officially open source, and the Agent is ready to install and use.

The intention of this series of combination punches is very obvious:

Alibaba Cloud does not intend to only be in the business of selling model APIs. What it does is water, electricity and coal in the Agent era - you need to make Agent applications, from infrastructure (OpenClaw cloud service), to model selection (Coding Plan), to capability invocation (Bailian CLI), to Agent framework compatibility (Claude Code/OpenClaw/Hermes/Qoder), the entire link is covered.

This is exactly the same logic as AWS's S3+EC2+Lambda in the cloud era - to sell a single service, but also to sell the entire ecosystem.

What does it have to do with projects like OpenClaw and Hermes?

People who understand the Agent ecosystem will ask: Since Alibaba has its own Qwen Code and Bailian, why does it take the initiative to be compatible with third-party frameworks such as OpenClaw and Hermes?

The answer is very realistic: the developer's workflow has long been occupied by these open source Agent frameworks.

  • OpenClaw has 356,000 stars and is one of the fastest growing open source projects in the history of GitHub
  • Hermes Agent has 71,000 stars in more than two months, and independent developers are the most active in migrating
  • Claude Code and Qoder are leading AI programming tools at home and abroad.

Alibaba Cloud's choice is: instead of competing with these projects for users' minds, it is better to be a good infrastructure provider directly - I don't care what Agent framework you use, anyway, you can use my capabilities.

This is a very mature strategy for selling shovels. Gold diggers come and go, shovels are always on sale.

Practical advice for developers

illustration

If you want to try it today, go in this order:

Step One: Get Free Token

Go to the Bailian console to open an account, and each model will receive 1 million Tokens, with a maximum limit of 50 million. Ordinary developers can’t finish it in two months.

Step 2: Install CLI

npm install -g bailian-cli
bl auth login --api-key sk-xxxxx

Step 3: Test your most valuable abilities

Try:

  • bl image generate (image generation, compare MidJourney)
  • bl video generate (Video generation, HappyHorse 1.0)
  • bl omni (multimodal dialogue)
  • bl text chat (Qwen3.7-Max, compared to Claude)

Step 4: Connect to your Agent

If you are using Claude Code, OpenClaw, Hermes or Qoder, directly configure bl as an available tool, and the Agent can call it autonomously.

Step 5: Running Video Demo

Use the official spark-video Skill to try generating a short video in one sentence and feel what the complete Agent workflow is like.

write at the end

The competition in the AI Agent era is no longer about whose model has the highest score, but who can make it easier for the Agent to use multi-modal, multi-ability, and cross-tool composite capabilities.

The real weight of Bailian CLI open source is that it smoothes out the most painful link of calling capabilities. In the past, if you wanted to make an Agent that could read pictures, write articles, generate videos, search the Internet, and call knowledge bases, you had to connect to more than 5 APIs. Now it’s all done with a bl command.

For individual developers: write 80% less boilerplate code and focus on product logic.

For enterprise teams: It saves a bunch of operation and maintenance costs for API integration, and gets more domestic compliance endorsements from the Bailian platform.

For the entire Agent ecosystem: Bailian CLI’s atomization capability + Agent-friendly design model may be imitated by other major manufacturers. In the next year or two, we will most likely see Tencent Cloud, Huawei Cloud, and Byte Volcano Engine follow up with similar CLI tools.

AI Agent’s battle for water, electricity and coal has officially begun.

GitHub address: github.com/modelstudioai/cli