> ## Documentation Index
> Fetch the complete documentation index at: https://docs.valkyrie.vals.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# How Valkyrie works

> Run scalable, reproducible evaluations for AI agents.

Valkyrie orchestrates agentic benchmark runs across sandboxed tasks. Use the CLI or Python SDK to start runs, monitor progress, and retrieve results.

<CardGroup cols={2}>
  <Card title="Run your first benchmark" icon="rocket" href="/get-started/quickstart">
    Install the CLI, configure credentials, and start a run.
  </Card>

  <Card title="Manage agents" icon="package" href="/agents/manage-agents">
    Upload, list, download, and remove agent bundles.
  </Card>

  <Card title="Add a benchmark service" icon="plug" href="/benchmarks/custom-services">
    Connect Valkyrie to a custom benchmark endpoint.
  </Card>

  <Card title="Self-host Valkyrie" icon="server" href="/self-hosting/infrastructure">
    Review the AWS architecture and deployment requirements.
  </Card>
</CardGroup>

## Core concepts

* **Agent**: a bundled program with a contract that tells Valkyrie how to install and run it.
* **Benchmark service**: an HTTP service that provides tasks and evaluates results.
* **Run**: one execution of an agent against a benchmark dataset.
* **Sandbox provider**: the service that creates isolated task environments.

## Choose an interface

Use the CLI for interactive setup, agent packaging, and run operations. Use the [Python SDK](/sdk/quickstart) to integrate run management into an application.