> ## Documentation Index
> Fetch the complete documentation index at: https://litellmagentplatform-fix-document-harness-auth-token.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> Run agents in sandboxes.

Self-hosted Kubernetes infrastructure for Claude Code, Codex, and any coding agent — **without ever giving them your real API keys**. A vault proxy swaps stub credentials for real keys at the wire level, so agents never hold the real values.

<img src="https://mintcdn.com/litellmagentplatform-fix-document-harness-auth-token/-SgpcgCh64tbVJRQ/images/lap-demo.gif?s=e56008651bbd7941aff7173c93c5d878" alt="LiteLLM Agent Platform demo" width="964" height="720" data-path="images/lap-demo.gif" />

```bash theme={null}
$ lap claude-code-cli1
```

## How it works

<img src="https://mintcdn.com/litellmagentplatform-fix-document-harness-auth-token/-SgpcgCh64tbVJRQ/images/architecture.svg?fit=max&auto=format&n=-SgpcgCh64tbVJRQ&q=85&s=6011e213727dd539c9063abe2cbb1b72" alt="LiteLLM Agent Platform architecture: sandbox pod containing Claude Code, Codex, opencode, Agent SDK harnesses, with a vault sidecar that swaps stub credentials for real keys before outbound HTTPS" width="820" height="400" data-path="images/architecture.svg" />

The agent process only ever sees stub credentials. The vault sidecar swaps them for real keys at the wire level, so real values exist only in the sidecar's memory — never in the agent process, never logged, never stored.

## Get started

<CardGroup cols={3}>
  <Card title="Install" icon="bolt" href="/installation">
    Run locally on kind, or deploy to AWS EKS.
  </Card>

  <Card title="Run Claude Code" icon="robot" href="/quickstart/claude-code">
    Open your first sandbox from the terminal.
  </Card>

  <Card title="Learn the vault" icon="shield-halved" href="/learn/vault-proxy">
    How agents stay isolated from real credentials.
  </Card>
</CardGroup>