Skip to content
DigitalNeuron
Tools & products

AWS Adds Session Endpoints to Bedrock AgentCore Runtime for Live App Previews

AWS introduced Session endpoints in Amazon Bedrock AgentCore runtime, letting developers preview web apps in a browser while a coding agent builds them.

By DigitalNeuron Desk3 min read

Quick answer

What did AWS announce for previewing coding agent output in Amazon Bedrock AgentCore?

AWS said Amazon Bedrock AgentCore runtime now supports Session endpoints, authenticated URLs that let developers open a web app a coding agent is building inside an isolated session directly in their browser. The endpoints forward HTTP and WebSocket traffic to ports the developer allows, so changes appear without a separate deployment step.

Key takeaways

  • AWS introduced Session endpoints, a capability of Amazon Bedrock AgentCore runtime that lets developers open a web app running inside an agent's session directly in their browser.
  • Each agent session runs in its own isolated microVM, and developers choose which ports a Session endpoint can reach, between one and five entries within the range 8000-12000, by setting the allowedPorts parameter when calling CreateAgentRuntime.
  • AWS said every request through a Session endpoint is authenticated, using SigV4 signatures for runtimes that rely on IAM or bearer tokens for runtimes with a JWT authorizer.
  • AWS published a sample in the agentcore-samples GitHub repository showing a coding agent that keeps a Vite website running in its session so changes appear in the browser as the agent edits files.
  • AWS said a Session endpoint is scoped to a single session and intended for building, testing, and reviewing agent output, not for serving apps to end users, which requires deploying to a separate hosting service such as AWS Amplify Hosting.

What AWS announced

Amazon Web Services said Amazon Bedrock AgentCore runtime now supports a capability called Session endpoints, which lets developers open and interact with a web app that a coding agent is building, directly in their own browser, while the agent is still working on it.

According to AWS, AgentCore runtime already runs each agent session in its own isolated microVM. Session endpoints add an authenticated URL for that session that forwards each request's HTTP method, path, query string, and body, as well as WebSocket connections, to ports the developer has allowed. AWS said this lets a developer watch a page update as the agent saves each file, rather than waiting for the app to be copied out and deployed elsewhere before each review.

How Session endpoints work

AWS described six details that govern how Session endpoints function:

  • Port selection: Developers set an allowedPorts parameter when calling CreateAgentRuntime, specifying one to five entries, each a single port or a range within 8000-12000, excluding reserved platform ports. AWS noted that allowing a port does not start a server; the application must listen on it, bound to 0.0.0.0.
  • URL delivery: The InvokeAgentRuntime response carries the endpoint URL in the X-Amzn-Bedrock-AgentCore-Runtime-Session-Endpoint-Url header, which SDKs that support it expose as sessionEndpointUrl.
  • Authentication: For runtimes using AWS IAM, AWS said each HTTP request, HTTP/2 stream, and WebSocket handshake carries a Signature Version 4 signature, and the caller needs the bedrock-agentcore:InvokeAgentRuntimeProxy permission. For runtimes with a JWT authorizer, each request carries a bearer token instead.
  • Port routing via host name: AWS said the client always connects on port 443, and a prefix on the host name selects the target port inside the session.
  • Protocol support: AWS said the endpoints carry HTTP/1.1, HTTP/2 (including server-sent events and gRPC), and WebSocket traffic.
  • Session scope: AWS said the URL requires credentials on every request and lasts only as long as the session, and is meant for building, testing, and reviewing agent output. To serve an app to end users, AWS said developers should deploy it to a hosting service such as AWS Amplify Hosting.

AWS also said that because a browser cannot attach a SigV4 signature or bearer token on its own, the Session endpoint client in the bedrock-agentcore SDK can open a local endpoint on the developer's machine, which adds credentials to each request and WebSocket handshake before forwarding it to the browser.

AWS distinguished the Session endpoint from the runtime endpoint that AgentCore runtime already provides. A runtime endpoint, such as DEFAULT, controls which version of an agent callers reach and lets developers keep separate versions for development and production. A Session endpoint, by contrast, reflects whichever version of the agent is running in that specific session and stops working when the session ends.

The sample walkthrough

AWS published a sample with four parts: a coding agent running on port 8080 that edits a website's files using list, read, write, and edit tools; a Vite development server on port 8090 started by the agent, which serves the website and pushes saved changes to the browser over its hot-reload WebSocket; a local PreviewApp that serves a preview page, sends prompts to InvokeAgentRuntime, and opens a local endpoint for the website; and the preview page itself, which loads the website through the local and Session endpoints while sending prompts to the agent. The complete sample, including the notebook the post follows, is in the 14-remote-access folder of the agentcore-samples GitHub repository.

Source: AWS Machine Learning Blog, "Preview web apps live as your coding agent builds them with Amazon Bedrock AgentCore runtime," published October 30, 2026.

Frequently asked questions

What is a Session endpoint in Amazon Bedrock AgentCore runtime?
AWS describes it as an authenticated URL tied to a single agent session that forwards HTTP requests and WebSocket connections to ports the developer allows, letting a developer open an app the agent is building in their own browser.
How does AWS secure Session endpoint traffic?
AWS said every request is authenticated: runtimes using IAM require a SigV4 signature plus the bedrock-agentcore:InvokeAgentRuntimeProxy permission, while runtimes with a JWT authorizer require a bearer token on each request.
Can a Session endpoint be used to serve an app to end users?
No. AWS said the Session endpoint URL lasts only as long as the session and is meant for building, testing, and reviewing agent output; serving an app to users requires deploying it to a separate hosting service such as AWS Amplify Hosting.
Where can developers find sample code for this feature?
AWS pointed to the 14-remote-access folder of the agentcore-samples GitHub repository, which includes a notebook demonstrating a coding agent that keeps a Vite website running in its session for live preview.

Sources

  1. Preview web apps live as your coding agent builds them with Amazon Bedrock AgentCore runtime | Artificial Intelligence — Amazon Web Services (AWS)
Tagsawsamazon-bedrockagentcorecoding-agentsdeveloper-toolscloud-infrastructure

Related reading

AWS Details Multi-Turn Reinforcement Learning Tool on SageMaker AI

AWS published a technical guide describing Multi-Turn Reinforcement Learning (MTRL) on Amazon SageMaker AI, a serverless pipeline that trains models on multi-step agentic tasks using Amazon Bedrock AgentCore to host the agent and reward function, scoring entire action sequences rather than single prompts.

3 min read

AWS Brings ChatGPT Desktop App to Amazon Bedrock in GovCloud (US)

AWS says the ChatGPT desktop app, including OpenAI's Work and Codex, is now available through Amazon Bedrock in AWS GovCloud (US). Codex routes requests through Amazon Bedrock Mantle in us-gov-west-1, governed by AWS IAM. AWS says pricing is per-token, zero data retention is supported, and data stays within the selected Region.

2 min read

AWS details IskayPet's corporate knowledge assistant built on Amazon Bedrock Managed Knowledge Bases

AWS published a case study describing how IskayPet, a pet care retailer operating in Spain, Portugal and France, built a corporate knowledge assistant on Amazon Bedrock Managed Knowledge Bases. The assistant answers employee questions in four languages by retrieving and citing passages from IskayPet's existing SharePoint documents, using hybrid search, reranking, and document-level permissions.

2 min read