# Teleport Beams

> Teleport Beams provides trusted runtimes for AI agents. Each beam is a short-lived, isolated Cloud Hypervisor micro-VM with virtualized networking and snapshots. Teleport delegates the user's identity to the agent, so it can reach cloud infrastructure and inference without secrets or static API keys. Beams captures protocol-level audit and LLM traces, supports reviewed agentic access requests, and flags harmful agent behavior such as emergent coordination and goal hijacking. Beams is a Teleport project, launching November 9, 2026.

## Pages

- [Home](https://www.beams.run/index.md): Overview of access control, swarm observability, delegated identity and the micro-VM environment.
- [Beams for AWS](https://www.beams.run/aws.md): Amazon Bedrock inference, AWS account and resource auto-discovery, and IAM-based access with reviewed access requests.
- [How it works](https://www.beams.run/how-it-works.md): Scene-by-scene walkthrough of creating a beam, its architecture, lifecycle, connectivity and delegated identity.
- [Pricing](https://www.beams.run/pricing.md): Pro and Enterprise plans, included usage, overage rates and a cost calculator.

## Docs

- [Beams](https://goteleport.com/docs/beams.md): What beams are, how they work, identity delegation, inference endpoints and use cases.
- [Beams Database Demo](https://goteleport.com/docs/beams/databases.md): Give an AI agent restricted access to a Teleport-protected database through a beam.
- [Manage Long-Running Agents on Teleport Beams](https://goteleport.com/docs/beams/long-running-agents.md): Run and manage services on beams with beamctl.