agentarts-cli

name	agentarts-cli
description	Use when working with the AgentArts CLI — agentarts init, agentarts dev, agentarts launch, agentarts invoke, agentarts config, or agentarts_config.yaml. Covers deployment, local dev, invoking deployed agents, troubleshooting ARM64/Observability/Gateway issues, and the .agentarts_config.yaml schema.

AgentArts CLI

AgentArts is Huawei Cloud's enterprise-grade Agent lifecycle platform. The CLI (agentarts) manages the full pipeline: init → dev → deploy → invoke.

1. Installation & Auth

pip install agentarts-sdk

# Auth (required for launch and invoke)
export HUAWEICLOUD_SDK_AK="<your-ak>"
export HUAWEICLOUD_SDK_SK="<your-sk>"

2. CLI Commands Quick Reference

Command	Purpose
agentarts --version	Verify CLI installed
agentarts init -n <name> -t langgraph	Scaffold a new LangGraph project
agentarts dev	Start local dev server on port 8080
agentarts launch	Build image → push SWR → deploy to Runtime
agentarts invoke '{"message":"..."}'	Call deployed Agent (sync)
agentarts invoke --custom-path stream '{"message":"..."}'	Call with custom sub-path (e.g., SSE streaming)
agentarts config	Interactive config wizard
agentarts runtime describe	Inspect deployed Runtime status

Local dev verification

agentarts dev
curl -X POST http://localhost:8080/invocations \
  -H 'Content-Type: application/json' \
  -d '{"message":"你好"}'

Invoke deployed Agent

# Sync
agentarts invoke '{"message":"帮我查一下明天的日程"}'

# SSE stream (requires PREFIX_MATCH in invoke_config.url_match_type)
agentarts invoke --custom-path invocations/stream '{"message":"你好"}'

Runtime status check

agentarts runtime describe

3. Project Structure

my_agent/
├── .agentarts_config.yaml  # Deployment config (see §4)
├── agent.py                # Agent entrypoint (if using AgentArtsRuntimeApp)
├── Dockerfile              # Container image definition
└── requirements.txt        # Python deps

For this project, the actual config lives at: personal-assistant-service/.agentarts_config.yaml

4. .agentarts_config.yaml Schema

default_agent: personal-assistant
agents:
  personal-assistant:
    base:
      name: personal-assistant
      entrypoint: agent:app          # Or app.main:app (FastAPI)
      dependency_file: pyproject.toml
      region: cn-southwest-2         # Only supported region
      platform: linux/arm64          # MUST be ARM64
      language: python3
      base_image: ghcr.io/astral-sh/uv:python3.12-bookworm
      container_runtime: docker

    swr_config:
      organization: personal-assistant-org
      repository: agent_personal_assistant
      organization_auto_create: true
      repository_auto_create: true

    runtime:
      arch: arm64                    # Must match base.platform
      invoke_config:
        protocol: HTTP
        port: 8080
        url_match_type: PREFIX_MATCH  # ACCURATE_MATCH (default, only /invocations)
                                      # PREFIX_MATCH (needed for /invocations/stream etc.)
      network_config:
        network_mode: 
      
         
         
      
        
           
        
           
        
           
      
         
         []
      
          
           
          
           

      
          
           
          
           

Key Config Fields

Field	Must-Know
base.platform	MUST be linux/arm64
base.region	Only cn-southwest-2 supported
runtime.arch	Must match base.platform — ARM64 images on x86 nodes fail silently
invoke_config.url_match_type	PREFIX_MATCH needed for sub-paths like /invocations/stream or /invocations/playground/
invoke_config.port	Always 8080
environment_variables	Secrets (API keys) go here, NEVER commit to git
artifact_source.url	Full SWR path; deploy with manual docker build + docker push before agentarts launch

5. Deployment Workflow (High-Level)

flowchart LR
    A[Set HUAWEICLOUD_SDK_AK/SK] --> B[docker build ARM64 image]
    B --> C[docker push to SWR]
    C --> D[agentarts launch]
    D --> E[Smoke test: /ping, /invocations, SSE]

Step-by-step

1. Set auth env vars:

   export HUAWEICLOUD_SDK_AK="<ak>"
   export HUAWEICLOUD_SDK_SK="<sk>"
   

2. Build ARM64 image (on x86 Mac, use buildx + QEMU):

   export BUILDKIT_USE_OCI_MEDIA_TYPES=0  # Required for Docker 27+
   docker buildx build --platform linux/arm64 --load \
     -f personal-assistant-service/Dockerfile \
     -t swr.cn-southwest-2.myhuaweicloud.com/personal-assistant-org/agent_personal_assistant:latest \
     .
   

   If local machine is not ARM64 native, first:

   docker run --rm --privileged multiarch/qemu-user-static --reset -p yes
   

3. Push to SWR:

   # Login (zsh-compatible password generation)
   SWR_PASSWORD=$(printf "$HUAWEICLOUD_SDK_AK" | openssl dgst -binary -sha256 -hmac "$HUAWEICLOUD_SDK_SK" | od -An -vtx1 | tr -d ' \n')
   echo "$SWR_PASSWORD" | docker login swr.cn-southwest-2.myhuaweicloud.com -u cn-southwest-2@$(echo -n "$HUAWEICLOUD_SDK_AK" | tail -c 16) --password-stdin
   
   docker push swr.cn-southwest-2.myhuaweicloud.com/personal-assistant-org/agent_personal_assistant:latest
   

4. Deploy:

   cd personal-assistant-service
   agentarts launch
   

5. Verify (after launch, note the Runtime domain from output):

   curl -s "<runtime-domain>/ping"
   # Expected: {"status": "ok"}
   
   agentarts invoke 
   
   
   agentarts invoke --custom-path invocations/stream 
   
   

6. Common Pitfalls

ARM64 architecture

• AgentArts Runtime only supports linux/arm64. X86 images cause exec format error.
• Build on x86 machines requires docker buildx + QEMU emulation.
• runtime.arch must match base.platform — both arm64.

Docker 27+ OCI media types

• SWR does not support OCI format images (Docker 27+ default).
• Always set export BUILDKIT_USE_OCI_MEDIA_TYPES=0 before building.

Gateway routing (url_match_type)

• Default ACCURATE_MATCH only forwards /invocations exactly.
• SSE streaming at /invocations/stream or Chainlit at /invocations/playground/ require PREFIX_MATCH.
• If routes return 404 at runtime but work locally, check this field.

IAM sub-account permissions

• The AK/SK must have SWR FullAccess (for push/launch) and OBS FullAccess (for frontend uploads).
• If using a sub-account, grant these in IAM → User Groups → Authorization.

Entrypoint mismatch

• .agentarts_config.yaml entrypoint and Dockerfile CMD should agree on the actual app module.
• If using FastAPI with app.main:app, set entrypoint: "app.main:app".
• When using manual docker build + docker push + agentarts launch, the Dockerfile CMD wins.

CORS

• If frontend and backend are on different origins, CORS middleware must be configured in the backend (not in .agentarts_config.yaml).
• Set CORS_ALLOWED_ORIGINS env var in environment_variables, then read it in the app's CORS middleware.

Silent container failures

• agentarts invoke returning NA.30200: failed to start container with empty logs usually means runtime.arch mismatch.
• Run grep -A10 'runtime:' personal-assistant-service/.agentarts_config.yaml | grep 'arch:' to confirm.

7. Observability (Control Console)

After agentarts launch, verify in the web console at https://console.huaweicloud.com/agentarts/:

1. Runtime status: 智能体运行时 → confirm instance is "运行中"
2. Full-chain Trace: 观测 → 全链路 Trace → check recent traces
3. Metrics: 观测 → 指标监控 → QPS, latency, error rate
4. Logs: 观测 → 日志 → container stdout/stderr

8. Quick Health Checks

# CLI installed?
agentarts --version

# SWR reachable?
curl -s -o /dev/null -w '%{http_code}' https://swr.cn-southwest-2.myhuaweicloud.com/v2/
# Expected: 401 (auth required — means it's reachable)

# Auth set?
echo $HUAWEICLOUD_SDK_AK

# Config exists?
ls personal-assistant-service/.agentarts_config.yaml