cyicz123/autodl-instance-pro

AutoDL Instance Pro

Overview

Use this skill for AutoDL public cloud container instance Pro resources.

• API host: https://api.autodl.com
• Host override: AUTODL_PRO_HOST
• Token order: AUTODL_PRO_TOKEN, then fallback AUTODL_TOKEN, read only from this skill's local .env
• CLI: node <SKILL_DIR>/autodl-pro.mjs ...
• Important: Pro API access requires personal or enterprise verification.

Do not reuse Elastic private-cloud host/token variables. AUTODL_ELASTIC_HOST and AUTODL_ELASTIC_TOKEN belong to the sibling elastic deployment skill only.

Dry-run Response Checklist

When the user asks to show commands, review a payload, dry-run, or not execute, do not call AutoDL. The answer must say not executed and no live API call, then include:

• CLI command, such as node <SKILL_DIR>/autodl-pro.mjs create --json <config.json>
• Host: https://api.autodl.com, unless AUTODL_PRO_HOST overrides it
• Token namespace: AUTODL_PRO_TOKEN, fallback AUTODL_TOKEN
• Endpoint when known, such as POST /api/v1/dev/instance/pro/create, POST /api/v1/dev/instance/pro/power_on, or POST /api/v1/dev/instance/pro/image/save
• JSON payload/body for create, power-on, save-image, release, and other state-changing operations

Commands

node <SKILL_DIR>/autodl-pro.mjs create --json <config.json>
node <SKILL_DIR>/autodl-pro.mjs snapshot <instance_uuid>
node <SKILL_DIR>/autodl-pro.mjs status <instance_uuid>
node <SKILL_DIR>/autodl-pro.mjs list [--page-index 1] [--page-size 10]
node <SKILL_DIR>/autodl-pro.mjs power-on <instance_uuid> [--start-command "..."]
node <SKILL_DIR>/autodl-pro.mjs power-off <instance_uuid>
node <SKILL_DIR>/autodl-pro.mjs release <instance_uuid>
node <SKILL_DIR>/autodl-pro.mjs save-image <instance_uuid> --name <image_name>
node <SKILL_DIR>/autodl-pro.mjs list-images [--page-index 1] [--page-size 10]

Release Workflow

Release is destructive and the CLI does not silently power off first.

1. Check status.
2. Power off the instance.
3. Release the stopped instance.

node <SKILL_DIR>/autodl-pro.mjs status pro-xxxxxxxx
node <SKILL_DIR>/autodl-pro.mjs power-off pro-xxxxxxxx
node <SKILL_DIR>/autodl-pro.mjs release pro-xxxxxxxx

Operation Payloads

Power-on calls POST /api/v1/dev/instance/pro/power_on with:

{
  "instance_uuid": "pro-xxxxxxxx",
  "payload": "gpu",
  "start_command": "bash /root/start.sh"
}

payload is always "gpu". Omit start_command only when the user did not provide one.

Save-image calls POST /api/v1/dev/instance/pro/image/save with:

{
  "instance_uuid": "pro-xxxxxxxx",
  "image_name": "my-saved-image"
}

Create Config

Required fields:

| Field | Requirement | |---|---| | req_gpu_amount | Integer 1 through 4 | | expand_system_disk_by_gb | Integer 0 through 500 | | gpu_spec_uuid | GPU spec UUID | | image_uuid | Base/private image UUID | | cuda_v_from | Minimum CUDA version value, such as 118 |

Example:

{
  "req_gpu_amount": 1,
  "expand_system_disk_by_gb": 50,
  "gpu_spec_uuid": "GPU-RTX4090",
  "image_uuid": "image-xxxxxxxxxx",
  "cuda_v_from": 118
}

Reference IDs

GPU spec IDs vary by region and account availability. Confirm current IDs with AutoDL before creating paid resources.

| Example GPU | Example spec ID | |---|---| | RTX 4090 | GPU-RTX4090 | | RTX 3090 | GPU-RTX3090 | | A100 | GPU-A100 | | H100 | GPU-H100 |

Public base image UUID examples are illustrative; prefer listing private images or using the official console for exact current values.

| Example image | Example UUID | |---|---| | PyTorch CUDA 11.8 | image-pytorch-118-example | | Ubuntu 22.04 CUDA 12.2 | image-ubuntu-122-example |

Error Handling

The CLI prints structured JSON to stdout. On error:

{
  "status": "error",
  "error_type": "validation_error",
  "message": "Pro 创建参数有误",
  "details": {
    "errors": ["req_gpu_amount(5) 必须是 1 到 4"]
  }
}

For token_missing, write AUTODL_PRO_TOKEN in this skill's .env. For validation_error, fix the config used by node <SKILL_DIR>/autodl-pro.mjs create --json <config.json> before retrying; do not submit an invalid paid create request. For release, ask the user to confirm the target instance and power-off state before calling release.

References

• Full endpoint table: api-reference.md
• CLI examples: examples.md