ark-demo

A Helm chart for deploying model configurations to Ark (the Agentic Runtime for Kubernetes) for demo purposes.

This demo kit includes Models, Agents, Agents-as-Tools (sometimes called "Sub-agents"), MCP servers, and also includes Jupyter Notebooks showing different ways to call into Ark.

• Quickstart
• Resources Created
  • Models
  • Agents
  • Teams
  • Tools
  • MCP Servers
• Interactive Notebooks
• Demos

Quickstart

Install via Helm:

helm upgrade --install ark-demo oci://ghcr.io/dwmkerr/charts/ark-demo \
    --set models.anthropic.apiKey="your-key" \
    --set models.gemini.apiKey="your-key"

Or install locally:

cp custom-values.template.yaml custom-values.yaml
# Edit custom-values.yaml with your API keys
make install

Alternatively, set your API keys as environment variables:

# Set API keys for the providers you want to enable
export ANTHROPIC_API_KEY="your-anthropic-api-key-here"
export GEMINI_API_KEY="your-gemini-api-key-here" 
export AZURE_OPENAI_API_KEY="your-azure-openai-api-key-here"
export OPENAI_API_KEY="your-openai-api-key-here"

The chart automatically enables/disables providers based on which API keys you set as environment variables. You can customize the models and base URLs by editing custom-values.yaml if needed.

Default configuration includes:

• Anthropic: Claude models with API key from ANTHROPIC_API_KEY
• Gemini: Google Gemini models with API key from GEMINI_API_KEY
• Azure OpenAI: GPT models with API key from AZURE_OPENAI_API_KEY
• OpenAI: GPT models with API key from OPENAI_API_KEY

Each provider is automatically enabled only if its corresponding environment variable is set.

Then use the following commands to deploy the demo kit:

# Install the starter kit models on Ark
make install

# Uninstall the starter kit from Ark
make uninstall

# Check deployment status
make status

A set of models, agents, MCP servers, and tools will be created. Check them with:

kubectl get models
kubectl get agents
kubectl get mcpservers
kubectl get tools

Resources Created

The following resources are created:

Models

As long as the appropriate API Keys are provided, models for Azure OpenAI, Gemini, Anthropic and OpenAI providers will be created:

Agents

The following agents will be deployed:

Agent Name	Deployed When	Description	Tools
code-reviewer	Always	Code review specialist who helps improve code quality, identifies issues, and suggests improvements	None
planner	Always	Planning specialist who breaks down complex tasks into manageable steps and creates actionable plans	None
team-leader	Always	Team leader and strategic planner who coordinates work distribution among team members	None
lead-software-engineer	Always	Lead software engineer with access to development tools, GitHub operations, and best practices	Conditional: GitHub tools, AI Developer Guide tools, shell subagent
github-agent	mcpServers.github.enabled	GitHub specialist for repository management, issues, pull requests, workflows, and security	All GitHub MCP tools
shell-subagent	mcpServers.shell.enabled	Specialized shell operations agent that executes commands safely (used as agent-as-tool)	Shell MCP tool

Teams

Team Name	Strategy	Members	Description
coding-team	Sequential	planner, code-reviewer	Collaborative team with planner and code reviewer working sequentially on development tasks
engineering-team	Selector	planner, lead-software-engineer	Engineering team with planner and lead software engineer using AI-driven participant selection

Tools

The following custom tools are created:

Tool Name	Type	Deployed When	Description
call-shell-subagent	agent	mcpServers.shell.enabled	Executes shell commands via specialized shell operations agent (agents-as-tools pattern)

Many tools from each MCP server will also be discovered on startup.

MCP Servers

MCP servers can be enabled via Helm dependencies in custom-values.yaml:

Name	Values Field	Standalone Install
shell-mcp	mcpServers.shell.enabled	helm install shell-mcp oci://ghcr.io/dwmkerr/charts/shell-mcp
github-mcp	mcpServers.github.enabled	helm install github-mcp oci://ghcr.io/dwmkerr/charts/github-mcp --set github.token="your-token"
AI Developer Guide	mcpServers.aiDeveloperGuide.enabled	helm install ai-developer-guide-mcp oci://ghcr.io/dwmkerr/charts/ai-developer-guide-mcp

Set to true in custom-values.yaml and run make install (dependencies update automatically).

Interactive Notebooks

pip install jupyterlab

Command	Description
python -m jupyterlab notebooks/ark-openai-apis-basic.ipynb	Simple non-interactive demo showing basic ARK OpenAI API usage
python -m jupyterlab notebooks/ark-openai-apis.ipynb	Interactive interface with widgets for target selection and streaming modes

Demos

Demo	Description
PR Review Workflow	Argo workflow that reviews GitHub PRs using agents, with optional PR commenting and HTML summary reports