Agent Configuration allows you to customize AI engines, select models, choose execution environments, and create specialized agents for your automation needs. Agent Playground Overview

​
Agent Types

On Demand (Execute): Create custom agents with specific engines, tools, and configuration for immediate execution. Existing Agent from Organization: Use pre-configured agents from your organization’s catalog with inherited settings. Import Agent Preset: Load agent configurations from URLs, paste JSON/YAML directly, or generate new agents with AI assistance. Agent Playground Capabilities You can configure various capabilities for your agents including AI models, tools, and execution environments. Agent Playground Final Configuration

​
AI Engine & Model Configuration

​
Selecting AI Engines

Claude Code (Recommended): Anthropic Claude models with enhanced tooling capabilities LiteLLM: Universal LLM gateway supporting multiple providers Claude Code: Direct access to Claude models with advanced reasoning Execution Runner Selection

​
Available Models

Model Selection Dropdown Claude Sonnet 4: Latest Claude model with enhanced reasoning and performance Claude Opus 4: Most capable Claude model for complex analysis and problem-solving
GPT-4o: OpenAI’s most advanced multimodal model LiteLLM GPT-4o Configuration

​
Execution Runners

Choose from available execution environments:
  • kubiya-prod: Production environment
  • kubiya-hosted: Managed cloud runner (recommended)
  • core-testing-1, core-testing-2: Testing environments
  • costa-least-privileges: Security-focused environment
  • enforcer: Compliance-focused runner
  • gcp-no-vcluster: Google Cloud environment
  • omer-runner, pai-runner: Specialized runners
  • pargevestaging: Staging environment

​
Importing Agent Presets

​
From URL

Import Agent from URL Load agent configurations directly from GitHub, GitLab, or any public repository. Supports JSON, YAML, and YML formats.

​
Paste Configuration

Import Agent Paste Config Paste JSON or YAML agent configurations directly into the interface for quick setup.

​
Generate with AI

Generate with AI Description Describe your desired agent functionality and let AI generate a complete configuration. Example prompts:
  • “agent that can interact with kubernetes clusters”
  • “I need an agent that can talk with AWS and create instances via EC2 running on hosted runner”
AI Agent Generation Process The AI assistant analyzes your requirements and generates a comprehensive agent configuration with:
  • Relevant tool searches
  • Agent detail extraction
  • Customized capabilities

​
Generated Agent Configuration

Generated Kubernetes Agent Generated agents include:
  • Agent Name: Descriptive name (e.g., “Kubernetes Operations Manager”)
  • Description: Comprehensive capabilities overview
  • AI Instructions: Detailed operational guidelines
  • Tools: Pre-configured tool set (kubectl-cluster-info, etc.)
  • Environment Variables: Required configuration (KUBECONFIG, KUBECTL_VERSION)
Agent Configuration Details Review the complete agent specification including:
  • Core capabilities and operational guidelines
  • Tool configurations and Docker images
  • Environment variables and secrets
  • JSON specification for export

​
Agent Preview & Import

Agent Preview and Import Before importing, review the agent configuration showing:
  • Tools: Number of included tools (3)
  • Secrets: Required secrets count (0)
  • Integrations: Connected services (0)
  • Environment Variables: Configuration variables (2)
Actions available:
  • Accept & Use Agent: Import and make available
  • Reject: Cancel the import
  • Continue Chatting: Modify the configuration

​
Using Configured Agents

Final Task Interface with Imported Agent Once configured, your custom agents appear in the task creation interface alongside the AI engine and execution environment, ready for workflow automation.

​
Configuration Best Practices

Model Selection:
  • Use Claude Sonnet 4 for most general-purpose tasks
  • Choose Claude Opus 4 for complex reasoning and analysis
  • Select GPT-4o for multimodal tasks requiring vision capabilities
Runner Selection:
  • Use kubiya-hosted for production workloads (managed and recommended)
  • Choose specialized runners for specific security or compliance requirements
  • Test configurations in staging environments before production deployment
Agent Design:
  • Create focused agents with specific domain expertise
  • Include comprehensive AI instructions for consistent behavior
  • Configure only necessary tools to maintain security boundaries
  • Use descriptive names and documentation for team collaboration
Security Considerations:
  • Review generated configurations before importing
  • Validate tool permissions and access scopes
  • Use least-privilege execution environments
  • Regularly audit agent capabilities and usage
Quick start: Begin with the “Generate with AI” option to create your first custom agent, then iterate on the configuration based on your workflow needs.