Ark CLI

npm i -g @volcengine/ark-cli@latest

What Ark CLI Can Do

Give your AI agent these Ark capabilities and let it compose them for complex tasks:

• Model invocation: text dialogue, multimodal input, streaming output, multi-turn continuation
• Image / video generation: Seedream image generation, Seedance video generation, with outputs automatically downloaded locally
• Multimodal understanding: OCR, video summarization, speech transcription, subtitle timing, document field extraction
• Plan management: Agent Plan / Coding Plan lookup, purchase, renewal, and team seat management
• Agent integration: install Ark Skills, configure Harness / MCP, and let local agents invoke Ark capabilities
• Model production: fine-tuning jobs, custom models, and training status tracking
• Service launch: one-click deployment of base or custom models as Endpoints
• Governance loop: Usage, Billing, Doctor, monitoring metrics, and error-code diagnostics

Why You Need Ark CLI

Ark CLI isn't about memorizing more commands — it turns Volcano Ark's MaaS platform capabilities into a toolbox that your agent can call.

消费链路:
选择模型
  -> 开通 Plan 或部署模型
  -> 接入本地 Agent / Harness
  -> 调用、理解、生成
  -> 查看用量、账单、监控和诊断结果

生产链路:
准备数据
  -> 创建精调任务
  -> 得到自定义模型
  -> 部署成 Endpoint
  -> 回到消费链路继续使用

Install & Init

# 1. 安装 CLI
npm i -g @volcengine/ark-cli@latest
arkcli --version

# 2. 登录火山方舟
arkcli auth login volc-sso
arkcli auth status

# 3. 同步 Ark Skills 到本地 Agent
arkcli +connect

For headless environments (e.g., remote dev machines, CI, or agent sandboxes):

arkcli auth login --no-browser
arkcli auth login --no-browser --code <授权码>

A Complete Task in One Go

Just give your agent a goal:

帮我选择一个适合图文理解的模型，配置到本地 Agent，
然后用它总结一段视频，并告诉我本月用了多少 token。

The agent can orchestrate the full pipeline through Ark CLI:

arkcli models search
  -> arkcli plans get / arkcli plans model-list
  -> arkcli +connect / arkcli helper configure
  -> arkcli +understand video-summary
  -> arkcli usage stats / arkcli billing list

You describe the goal; the agent picks the commands, calls Ark, and aggregates results.

Common Tasks

Select & Call a Model

arkcli +chat "总结一下 RAG 的核心思想"

arkcli +chat --input @photo.jpg "描述这张图片"

arkcli +understand video-summary --input @demo.mp4 "按章节总结这个视频"

arkcli +gen "生成一张未来城市海报"

Manage Plans

# 查看当前账号持有哪些套餐
arkcli plans get

# 查看 Plan 支持哪些模型
arkcli plans model-list --plan agent-plan

# 查看团队版席位
arkcli plans team seat-list --plan agent-plan-team

# 查看本机 Agent 上 Plan 相关能力状态
arkcli plans harness-status

# 购买或续费前先查看参数
arkcli plans buy --help
arkcli plans renew --help

plans buy / plans renew are billing operations. Ark CLI will show the agreement and a confirmation step first — it will not charge you by default.

Connect Ark Service Capabilities

+connect installs Ark Skills so the agent knows what Ark tasks are available. helper configures Harness / MCP / Plan, giving the agent real access to models and tooling.

# 查看会检测到哪些本地 Agent
arkcli +connect list

# 安装 Ark Skills
arkcli +connect

# 清理已注入的 Ark Skills
arkcli +connect uninstall

# 查看 Harness / Agent 配置状态
arkcli helper list

# 给当前 Agent 注入 MCP 工具能力
arkcli helper mcp

# 配置指定 Harness，并同步注入 MCP
arkcli helper configure codex --with-mcp

arkcli +connect uninstall removes the ark- / arkcli- prefixed Skills injected into your local agent. Use it when you want a clean agent environment or are concerned that too many Skills may affect agent routing. It does not uninstall the arkcli command itself.

Production Models

arkcli train finetune --help
arkcli train finetune capability get --model <model> --version <version>
arkcli train finetune create --help
arkcli train finetune watch <job-id>

Deployed Services

arkcli +deploy --name my-endpoint --model doubao-seed-2-0-pro-260215 --dry-run
arkcli +deploy --name my-endpoint --model doubao-seed-2-0-pro-260215
arkcli infer endpoint list
arkcli infer endpoint get <endpoint-id>

Manage Costs, Status, and Health

arkcli usage plan
arkcli usage stats --start 2026-07-01 --mine
arkcli billing list --start 2026-07
arkcli billing list --start 2026-07 --endpoint ep-xxx

arkcli doctor
arkcli doctor model <model-name>
arkcli doctor infer-endpoint <endpoint-id>
arkcli doctor metrics request.qpm
arkcli doctor error ContentRiskBlocked

By default, doctor is read-only and won't modify resources. Fixes require confirmation.

Commands

Each command has its own help:

arkcli +chat --help
arkcli +gen --help
arkcli +understand --help
arkcli +deploy --help
arkcli plans --help
arkcli doctor --help

Auth

Volcengine SSO is recommended:

arkcli auth login volc-sso

After signing in, Ark CLI automatically completes identity, profile, and API Key configuration. You usually don't need to manually copy keys.

arkcli auth status
arkcli auth apikey
arkcli auth logout

Config

# 查看当前 profile
arkcli profile show

# 列出所有 profile
arkcli profile list

# 切换 profile
arkcli profile use <profile-name>

For more profile management capabilities, see arkcli profile --help.

AI Agent Skills

The skills/ directory in this repository collects the built-in Agent Skills for Ark CLI. Run:

arkcli +connect

Ark CLI installs these Skills onto your local Agent, enabling the Agent to invoke Volcengine Ark capabilities through natural language, including chat, multimodal understanding, image and video generation, model lookup, Plan management, Endpoint deployment, fine-tuning, usage, billing, and diagnostics.

User Community

Scan the QR code to join the Ark CLI Feishu user community for usage support, troubleshooting, bug feedback, and experience sharing. You can ask questions about installation, login, model invocation, Skill usage, and version upgrades. Feedback on issues not covered by GitHub Issues is also welcome.

Please do not share API Keys, AK/SK, Tokens, or other sensitive information in the group. For bugs that can be stably reproduced, it is recommended to also file a GitHub Issue to facilitate tracking and fixes.

Links

• Volcengine Ark: https://www.volcengine.com/product/ark
• npm: https://www.npmjs.com/package/@volcengine/ark-cli
• GitHub Releases: https://github.com/volcengine/ark-cli/releases
• Built-in Skills: https://github.com/volcengine/ark-cli/tree/main/skills

Changelog

Release notes are available in GitHub Releases.

License

Apache-2.0