Agent Ops

Demos, guides, and getting-started material for running OpenShell on OpenShift.

Guides

Getting Started with OpenShell on OpenShift

End-to-end guide for installing OpenShell with Helm, exposing the gateway through an OpenShift Route, configuring mTLS, registering a provider, creating a sandbox, running Claude Code in the sandbox, and managing egress policies.

Running OpenShell sandboxes with Kata runtime on OpenShift

Configure OpenShell sandboxes to use a Kata-backed RuntimeClass in sidecar topology on OpenShift, then verify the VM isolation boundary and network policy enforcement.

Inference Routing with RHOAI

Route sandbox inference traffic through a token-authenticated RHOAI-served model using the OpenShell privacy router, without exposing credentials to the sandbox.

OpenShell Capability Testing and Security Analysis

Testing results for OpenShell v0.0.85 on OpenShift, including capability behavior in the supervisor and sandbox user contexts.

Demos

MLflow OpenShell Tracing

Demonstrates how to capture MLflow traces from AI agents running in OpenShell sandboxes and send them to the managed MLflow instance on RHOAI.

The demo includes:

• MLflow auto-instrumentation — mlflow.openai.autolog() captures all LLM calls as traces with zero code changes
• OpenShell inference routing - Agent code sends requests to inference.local through the OpenAI SDK, and the OpenShell proxy handles model credentials
• Environment variable injection — Passes MLFLOW_TRACKING_URI by using --env instead of --credential for direct SDK access
• Sandbox network policy — Configures explicit network access from sandboxed workloads to the MLflow tracking server

Stack: Python, OpenAI SDK, MLflow, OpenShell, RHOAI

See the MLflow OpenShell Tracing README for setup and usage.