Skip to content

Stack Updates & Upgrades

This guide covers how to update an existing LMA stack to a new version or change its configuration. Stack updates allow you to adopt new features, apply bug fixes, and adjust parameters without deleting and recreating the stack.

The simplest way to update your stack:

Terminal window
# Update from the latest published template for your region
lma-cli deploy --stack-name LMA --wait
# Or update from local code changes
lma-cli deploy --stack-name LMA --from-code . --wait

The CLI auto-selects the correct template for your region and streams deployment events in real-time.

  1. Log into the AWS Console.
  2. Navigate to CloudFormation and select your LMA stack.
  3. Choose Update and then Replace current template.
  4. Enter the template URL for your region (see table below).
  5. Review and adjust parameters as needed.
  6. Click Next twice, check the IAM acknowledgement boxes, and click Update stack.
RegionTemplate URL
US East (N. Virginia)https://s3.us-east-1.amazonaws.com/aws-ml-blog-us-east-1/artifacts/lma/lma-main.yaml
US West (Oregon)https://s3.us-west-2.amazonaws.com/aws-ml-blog-us-west-2/artifacts/lma/lma-main.yaml
Asia Pacific (Tokyo)https://s3.ap-northeast-1.amazonaws.com/aws-bigdata-blog-replica-ap-northeast-1/artifacts/lma/lma-main.yaml
Europe (Ireland)https://s3.eu-west-1.amazonaws.com/aws-bigdata-blog-replica-eu-west-1/artifacts/lma/lma-main.yaml

If you are building from source, use lma-cli deploy --from-code . to build and deploy in one step, or use lma-cli publish to build first and then deploy with the resulting template URL.

Terminal window
# Build and deploy in one step
lma-cli deploy --stack-name LMA --from-code . --wait
# Or publish first, then deploy separately
lma-cli publish --source-dir . --region us-east-1
lma-cli deploy --stack-name LMA --template-url <url-from-publish-output> --wait

The legacy publish.sh script is also still available — use its template URL output with lma-cli deploy --template-url <url>.

The following data and configuration are preserved when you update or upgrade your stack:

  • Custom prompt templates stored in DynamoDB
  • User data including meetings, transcripts, and recordings
  • Cognito user accounts (v0.2.0 and later)
  • Installed MCP servers

The following may be modified during an update:

  • Default prompt templates
  • Infrastructure resources (Lambda functions, ECS tasks, etc.)
  • Lambda function code

The EnableDataRetentionOnDelete parameter (default true) decides what happens to the resources that hold durable data when the stack itself is deleted. With it set to true, deleting the stack leaves the following behind in your account rather than removing them; the resources it deliberately does not cover are listed after the table.

ResourceStackContents
EventSourcingTableAI stackMeetings, transcripts and summaries
VirtualParticipantTableAI stackVirtual participant records and schedules
MCPServersTableAI stackInstalled MCP server configurations
MCPApiKeysTableAI stackHashed per-user MCP API keys
OAuthStateTableAI stackIn-flight OAuth authorization state
VPTaskRegistryAI stackVirtual participant task registry
DomSelectorCacheAI stackCached meeting-platform UI selectors
VPProfilesBucketAI stackVirtual participant browser profiles
CallEventProcessorDiscardedRecordsQueueAI stackTranscript records the pipeline did not apply, awaiting triage
RecordingsBucketMain stackMeeting audio and video recordings, and transcript files
LoggingBucketMain stackS3 server access logs and load balancer logs
CustomerManagedEncryptionKeyMain stackThe KMS key every other retained resource is encrypted with
UserPoolCognito stackUser accounts and group memberships
IdentityPoolCognito stackIdentity pool and the identity ids issued from it
LLMPromptTemplateTableLLM template stackPrompt templates edited from the web UI
ChatButtonConfigTableChat button stackChat button definitions edited from the web UI
NovaSonicConfigTableNova Sonic stackVoice assistant prompt and model settings
AsrConfigTableASR MicroVM stackDiarization operating point tuned from the ASR Config page
S3VectorBucket, S3VectorIndexBedrock KB stackKnowledge base vector store and its embeddings

CustomerManagedEncryptionKey matters more than it looks: without the key, the retained tables and buckets cannot be read. Keep it for as long as you keep anything encrypted with it.

Retained resources continue to incur storage charges and must be removed by hand when you no longer need them — see Cleanup. Set EnableDataRetentionOnDelete to false if you would rather a stack deletion remove everything.

The parameter governs everything that holds durable user data. These resources are deleted with the stack either way, deliberately:

ResourceStackWhy it is not retained
WebAppBucketAI stackBuild artifacts a redeployment regenerates; emptied on deletion regardless of this parameter
MCPServerExternalAppClientAI stackConfiguration rather than data; recreated by a redeployment
TranscriberCallEventTableTranscriber stackShort-lived, TTL’d handoff state for the Post Call Analytics path, not user data. Its template does not read this parameter at all
AsrImageSourceBucketASR MicroVM stackPer-stack build context for the MicroVM image, rebuilt from the repository on the next deployment
CustomHeaderNameSecret, CustomHeaderValueSecretTranscriber stackCloudFront origin-verification secrets, regenerated per deployment; retaining them would leave Secrets Manager entries that a new stack cannot use
CategorySNSTopicAI stackA topic, not a store — it holds no messages between deliveries. Any subscriptions you added to it by hand are lost with it

One consequence to be aware of when deleting a stack: AsrImageSourceBucket has versioning enabled and, unlike WebAppBucket, no custom resource that empties it first, so a stack deletion can fail on a non-empty bucket. Empty it by hand and retry the deletion if that happens.

The retention setting is stored as a CloudFormation resource attribute, so changing the parameter on an existing stack only takes effect once the stack update that carries it has completed.

This is a major consolidation release. The following features and integrations have been removed:

  • QnABot
  • Amazon Lex
  • Bedrock Agent
  • Amazon Q Business
  • OpenSearch
  • Chrome browser extension
  • Healthcare domain support
  • Claude 3.x models (Claude 4+ only)

Only the Strands Bedrock agent remains as the meeting assistant service. If you relied on any of the removed features, they will no longer be available after upgrading to v0.3.0.

Breaking change: The Cognito user pool is deleted and recreated during this upgrade. All existing users must re-register after the update completes. See User-Based Access Control for details on user management.