agenttier
AgentTier is a Kubernetes-native runtime designed to securely run AI coding agents in isolated sandboxes. Each agent operates in its own Pod with a dedicated Persistent Volume Claim (PVC), ensuring complete separation of state and resources. The platform enforces strict default-deny NetworkPolicy to prevent unauthorized network access, and provides a streaming Server-Sent Events (SSE) invoke API for real-time communication with agents. It supports running multiple agents in parallel, making it suitable for orchestrating concurrent coding tasks at scale. Built with a focus on extensibility, AgentTier includes a Python SDK that allows developers to programmatically create, manage, and interact with agent sandboxes. A web UI is also available for visual management and monitoring. The system is packaged with a Helm chart and Terraform configurations, enabling straightforward deployment on any Kubernetes cluster. The core runtime is written in Go, ensuring high performance and efficient resource utilization. AgentTier addresses the growing need for robust agent orchestration in AI-driven development workflows. By leveraging Kubernetes primitives like Pods, PVCs, and NetworkPolicies, it provides enterprise-grade security and scalability. The default-deny network stance minimizes attack surfaces, while the sandboxed PVCs prevent cross-contamination between agents. The SSE API enables responsive, event-driven control without polling. Overall, AgentTier is an open source platform that bridges the gap between AI coding agents and production-grade infrastructure. It simplifies the deployment and management of agent fleets, making it an ideal foundation for teams building internal tools, automated code generation pipelines, or multi-tenant agent services.
- Category
- AI Agents
- Pricing
- Open Source
Problem solved
Development teams and organizations face challenges when deploying multiple AI coding agents concurrently: agents may interfere with each other, share sensitive state, or consume disproportionate resources. AgentTier solves this by providing a Kubernetes-native sandboxing layer that isolates each agent in its own Pod with a dedicated PVC and default-deny network policies. This ensures secure, conflict-free parallel execution without manual configuration. Additionally, the streaming API and SDK simplify integration into automated workflows, enabling reliable, scalable agent orchestration.
Use cases
- Run multiple Claude Code instances simultaneously on different coding tasks without cross-contamination
- Securely execute untrusted or third-party AI coding agents in a sandboxed environment
- Implement a CI/CD pipeline where each commit triggers an isolated agent to generate code or review changes
- Host a multi-tenant service that allows different users to run their own coding agents with guaranteed isolation
- Batch process large numbers of code generation requests in parallel, each in its own secure sandbox
Audience
- Developers
- ML Engineers
Pros
- Strong security isolation via per-agent Pods and default-deny network policies
- Kubernetes-native design ensures scalability and compatibility with existing infrastructure
- Supports parallel execution of multiple agents out of the box
- Includes a Python SDK and web UI for easy management and automation
- Open source with Helm and Terraform deployment options for production use
Cons
- Requires a Kubernetes cluster, which may be a barrier for smaller teams or simpler use cases
- Relatively new project with limited community and third-party integrations
Official website: https://github.com/agenttier/agenttier
Canonical LambdaBase page: https://www.lambdabase.com/tools/agenttier