GKE Agent Runtime

Name: GKE Agent Runtime
Availability: InStock
Author: Envision

Deploy sandboxed, multi-agent, and memory-enabled AI agents on GKE Kubernetes clusters.

1|Updated May 17, 2026

Search Tags:#ai agents #multi-agent #kubernetes #sandbox #adk #gke #vertex ai

Ready to run

npx skills add https://github.com/Envision-Construction/Envision-Skill-Repo --skill gke-agent-runtime-envision-construction

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: GKE Agent Runtime
Source: https://github.com/Envision-Construction/Envision-Skill-Repo/tree/main/plugins/infra/gke/skills/agent-runtime
Command: npx skills add https://github.com/Envision-Construction/Envision-Skill-Repo --skill gke-agent-runtime-envision-construction

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running AI agents in production requires isolated execution environments, orchestration of multi-agent workflows, and persistent memory across sessions, all of which are complex to set up manually on Kubernetes. ## Core Features & Use Cases - K8s-Native Agent Sandboxing: Deploy isolated agent pods using Sandbox, SandboxWarmPool, SandboxClaim, and SandboxTemplate CRDs with network isolation and resource limits. - 5-Agent Autonomous Pipeline: Implement the speccer → planner → builder → workers → reviewer pattern from Google's AI Factory, with spec/plan templates and reviewer guard sub-agents. - ADK Agents with Memory: Deploy Google ADK agents to GKE with Vertex AI Sessions for short-term state and Memory Bank for long-term recall. - Use Case: A platform engineer needs to run parallel coding agents on GKE. They create a SandboxWarmPool, issue a SandboxClaim per task, and collect artifacts via shared storage, achieving near-instant agent startup. ## Quick Start Ask the assistant to set up an agent sandbox warm pool on the GKE cluster in project claude-mcp-457317 and show how to claim sandboxes for individual tasks.

Frequently Asked Questions about GKE Agent Runtime

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I run AI agents in isolated sandboxes on GKE?▼

Install cert-manager and the kubernetes-sigs/agent-sandbox controller, then create Sandbox resources defining pod templates with resource limits. Use SandboxWarmPool for pre-warmed pods and SandboxClaim to request sandboxes from the pool on demand.

How do I deploy a Google ADK agent to GKE with memory?▼

Register an Agent Engine with memory_bank_config, then run adk deploy gke with session_service_uri and memory_service_uri pointing to agentengine://<engine-id>. Configure Workload Identity with the aiplatform.user role for cluster access to Vertex AI.

What is the difference between SandboxWarmPool and SandboxClaim?▼

SandboxWarmPool maintains a set of pre-warmed agent pods for near-instant startup, following a PVC/PV-like pattern. SandboxClaim requests a sandbox from a named warm pool and injects task-specific environment variables.

Can I use Claude models with the AI Factory multi-agent pipeline?▼

Yes, the speccer-planner-builder-worker-reviewer cycle is model-agnostic. Change the model field in each agent definition to claude-opus-4-7 or claude-sonnet-4-6, replace the Gemini Service Portal with ANTHROPIC_API_KEY, and use Claude Code's Agent tool for worker dispatch.

What is the difference between PreloadMemoryTool and LoadMemoryTool in ADK?▼

PreloadMemoryTool automatically fetches relevant memories on every turn via system instructions. LoadMemoryTool lets the agent decide when to retrieve memories through an on-demand tool call, giving more control over context usage.

When should I use VertexAiSessionService instead of DatabaseSessionService?▼

VertexAiSessionService is a managed service suited for production deployments without operating your own database. DatabaseSessionService stores sessions in your own SQL database for self-hosted production, while InMemorySessionService is only for development and testing.