Skip to Content
Introduction

Ark - Agentic Runtime for Kubernetes

ARK is an Kubernetes-based platform that provides a runtime and enables management and orchestration of AI agents and their resources. It provides custom resource definitions (CRDs) and services to handle:

  • AI model configurations
  • Autonomous agents
  • Multi-agent teams
  • Tool integrations
  • Memory persistence
  • Performance evaluation

Follow the Quickstart guide to install Ark locally.

Note

The Agents at Scale - Agentic platform and runtime for Kubernetes (“Ark”) Technical Preview should not be run without reading the Disclaimer, which covers key risk topics.

We are transitioning this project to our open-source repository. During this process we will be moving documentation URLs, artifact registries, etc. Some builds, artifacts and pipelines may be impacted.

Build, deploy and operate agentic solutions using proven patterns and practices that are reliable for complex distributed systems.

Key Features

  • Platform-agnostic agent operations
  • Standardized deployment patterns
  • Transparent orchestration
  • Built-in evaluation capabilities
  • Extensible tool integration
  • Multi-agent coordination
  • Custom headers support (for e.g. request tracing, custom metadata)

Architecture

High-level architecture diagram

Core Resources

  • Models: Configure and connect to AI models.
  • Agents: Create autonomous AI agents with specific capabilities and tools
  • Teams: Orchestrate multiple agents working together with coordination strategies
  • Queries: Execute prompts and manage conversations with agents or teams
  • Tools: Define custom tools and MCP tool references for agents
  • MCPServers: Configure Model Context Protocol servers for external integrations
  • Memory: Persistent storage for agent conversations and state
  • Evaluator: Services for assessing and scoring agent performance
  • Evaluation: Define evaluation configurations and results
  • ExecutionEngine: Specialized runtimes for different agent frameworks
  • A2AServer: Agent-to-agent communication services

MCP Servers

  • filesystem-mcp: File system operations and management

Services

  • ark-api: REST API for managing ARK resources (includes A2A Gateway for agent-to-agent communication)
  • ark-dashboard: Web-based management interface
  • ark-mcp: MCP server integration service
  • ark-evaluator: Holistic evaluation and scoring service
  • localhost-gateway: Local development gateway (only on local)
  • ark-cluster-memory: Memory storage with streaming support

Sample Services

  • langfuse: Observability and tracing service
  • executor-langchain: LangChain agent execution engine

Credits

The initial design and implementation of Ark was led by Roman Galeev , Dave Kerr , and Chris Madden .

Last updated on
Quickstart