Add Amazon Bedrock AgentCore Runtime deployment example and docs references

[sundargthb]

Summary

Add AgentCore Runtime as a deployment option for LiveKit agents. This involves:

1. A new bedrock-agentcore/ folder in livekit-examples/agent-deployment (PR submitted separately)
2. Two small documentation updates (described below) to link to it

AgentCore Runtime is AWS's serverless container hosting service for AI agents. We validated that a standard LiveKit agent runs on it unchanged — the agent joins LiveKit rooms as a normal WebRTC participant, just hosted inside an AgentCore container.

What was validated

• Agent packaged as ARM64 Docker container, deployed to AgentCore Runtime in VPC mode
• VPC configured with private subnets + NAT Gateway + Internet Gateway for UDP egress
• TURN server required for NAT traversal (two options work: KVS Managed TURN via AWS, or third-party like Cloudflare/Twilio)
• Agent connects to LiveKit Cloud as a standard WebRTC participant — no code changes to livekit-agents
• Works with the existing AWS plugin (Bedrock LLM, Polly TTS, Transcribe STT, Nova Sonic)

PR: Example code

PR submitted to livekit-examples/agent-deployment adding:

bedrock-agentcore/
├── README.md           # Full guide: VPC, TURN (KVS + third-party), Dockerfile, deploy script
├── agent.py            # Standard LiveKit agent with TURN configuration
├── kvs_turn.py         # KVS Managed TURN helper (GetIceServerConfig flow)
├── deploy.py           # boto3 script to create AgentCore Runtime
├── Dockerfile          # ARM64 container
├── requirements.txt
├── .env.example
└── .dockerignore

This follows the existing pattern (aws-ecs/, kubernetes/, render/).

Requested documentation updates

Update 1: Add to "Where to deploy" on Self-hosted deployments page

Page: /deploy/custom/deployments

Section: "Where to deploy" (the bullet list that currently shows LiveKit Cloud, Kubernetes, Render, More deployment examples)

Add this bullet after Render, before "More deployment examples":

- **[Amazon Bedrock AgentCore Runtime](https://github.com/livekit-examples/agent-deployment/tree/main/bedrock-agentcore)**: Serverless container hosting on AWS with auto-scaling and session isolation. Requires VPC mode with NAT Gateway and a TURN server for WebRTC media transport.

This matches the existing format — a linked name, brief description, and the link points to the example folder in agent-deployment.

Update 2: Add cross-link on AWS integration page

Page: /agents/integrations/aws

Location: At the bottom, after the "AWS plugin documentation" section

Add this section:

## Deploying to AWS

In addition to the Bedrock, Polly, Transcribe, and Nova Sonic plugins above, you can host your LiveKit agent on AWS using Amazon Bedrock AgentCore Runtime — a serverless container runtime with auto-scaling and session isolation. Your agent code stays the same; AgentCore just provides the hosting.

- **[AgentCore Runtime deployment guide](https://github.com/livekit-examples/agent-deployment/tree/main/bedrock-agentcore)**: Full setup instructions including VPC configuration, TURN server options (KVS Managed TURN and third-party), and deployment via the AWS SDK.

This gives AWS-focused developers a direct path from the AI services page to the deployment option, and makes it clear that the plugins and the deployment target are complementary.

Why this is complementary to LiveKit Cloud

• Expands deployment choice for AWS-centric teams without lock-in
• No code changes to the LiveKit SDK — agent runs unchanged
• Complements the existing Bedrock/Polly/Transcribe/Nova Sonic plugins
• Serverless + session isolation fills a gap for variable AI workloads
• No maintenance overhead for the LiveKit team — community example following existing patterns (Kubernetes, Render, AWS ECS)

Reference

• AgentCore Runtime developer guide
• AgentCore VPC configuration
• KVS GetIceServerConfig API
• Pipecat AgentCore WebRTC example (reference implementation in another framework)