ark-cli
Local development CLI for ark - run, test, and debug AI agent pipelines without a Kubernetes cluster.
Uses the same YAML files as kubectl apply. No cluster required.
Full documentation at docs.arkonis.dev.
Install
go install github.com/arkonis-dev/ark-cli/cmd/ark@latest
Or download a pre-built binary from the releases page.
Quick start
# Run a pipeline locally with a real LLM
ark run my-team.yaml
# Run with mock responses (no API key needed)
ark run my-team.yaml --provider mock
# Validate YAML without running
ark validate my-team.yaml
# Stream step-by-step output
ark run my-team.yaml --watch
# Machine-readable output for CI
ark run my-team.yaml --output json
Commands
| Command |
Description |
ark run <file> |
Execute an ArkTeam pipeline locally |
ark run --provider mock |
Run with fake LLM responses |
ark run --dry-run |
Validate YAML and estimate token cost |
ark run --watch |
Stream step-by-step output inline |
ark run --no-mcp |
Skip MCP tool connections |
ark run --output json |
Machine-readable output for CI |
ark validate <file> |
Validate YAML structure and DAG |
ark operator install |
Install ark into a cluster via Helm |
ark operator upgrade |
Upgrade an existing ark installation |
ark trigger <team> |
Trigger a team pipeline in-cluster |
ark status |
Show status of running pipelines |
ark logs <flow> |
Tail logs from a running pipeline |
Providers
| Flag |
Description |
--provider auto |
Auto-detect from model name (default) |
--provider anthropic |
Force Anthropic Claude |
--provider openai |
Force OpenAI / compatible endpoint |
--provider mock |
Fake responses - no API key needed |
Set OPENAI_BASE_URL to use any OpenAI-compatible endpoint (Ollama, vLLM, etc.).
Example output
Step "research" [running] ──────────────────── 4.2s
Step "research" [done] 1,204 tokens 2.1s
└─ "Here are 3 key findings about Kubernetes operator patterns..."
Step "summarize" [done] 312 tokens 1.8s
└─ "• Operators extend the Kubernetes API via CRDs..."
Flow succeeded in 9.1s - total: 1,516 tokens
Contributing
See CONTRIBUTING.md.
Code of Conduct
This project follows the Contributor Covenant Code of Conduct.
License
Apache 2.0 - see LICENSE.