AgentTier¶
Kubernetes-native platform for isolated, persistent sandboxes — for humans and AI agents.
AgentTier gives humans and AI agents disposable, on-demand dev environments managed as Kubernetes CRDs. Each sandbox is a Pod + PVC + NetworkPolicy with its own persistent workspace, a full PTY terminal in the browser, and optional per-session credentials. The warm pod pool makes creation feel instant (~800 ms in our measurements vs ~10 s cold).
When to use it¶
- Give each AI agent (Claude Code, Cursor, Aider, a custom agent) a private, isolated environment with its own storage and credentials.
- Run untrusted AI-generated code safely with gVisor kernel isolation.
- Provide on-demand developer environments for your team without pets.
- Orchestrate multi-agent workflows where agents need to collaborate over in-cluster networking.
Get started¶
- Quickstart — from zero to a running sandbox.
- Installation — Helm chart, values, and production knobs.
- Templates — the YAML blueprints that describe each agent.
- Governance — cluster-wide and per-namespace policy enforcement.
- Port forwarding — exposing sandbox ports via Services and Ingresses.
Install in three commands¶
helm repo add agenttier https://agenttier.github.io/agenttier/charts
helm repo update
helm install agenttier agenttier/agenttier --namespace agenttier --create-namespace
Then create a sandbox from the bundled general-coding template and open it in
your browser. See the Quickstart.
License¶
Apache-2.0. Source at github.com/agenttier/agenttier.