Agent Ops

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

Guides

Getting Started with OpenShell on OpenShift

End-to-end walkthrough covering Helm installation, Route-based gateway exposure, mTLS setup, provider registration, sandbox creation, running Claude Code inside a sandbox, and egress policy management.

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.

Demos

MLflow OpenShell Tracing

Enablement content showing how to capture MLflow traces from AI agents running in OpenShell sandboxes into the managed MLflow instance on RHOAI.

What it shows:

• MLflow auto-instrumentation — mlflow.openai.autolog() captures all LLM calls as traces with zero code changes
• OpenShell inference routing — Agent code calls inference.local via the OpenAI SDK; the OpenShell proxy handles model credentials
• Environment variable injection — MLFLOW_TRACKING_URI passed via --env (not --credential) for direct SDK access
• Sandbox network policy — Explicit network access to the MLflow tracking server from sandboxed workloads

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

See the MLflow OpenShell Tracing README for setup and usage.