Skip to content

Tiers and targets

llms.txt lists every page for a coding agent.

One source tree, three tiers. LOOM_TIER selects one; every difference below is a value resolved from it, so light and production-ha build from the same composites. What has run where is on Validation.

lightproductionproduction-ha
Intended useBrowse, evaluate, single-account devAdoptable single-AZ deploymentAdoptable high-availability deployment
VPCBuilt by the stack (public subnets, no NAT)YoursYours
PrivateLink and private subnetsOmittedProvisionedProvisioned
ECS AssignPublicIpENABLED, the only route to ECR without a NATDISABLEDDISABLED
ServingHTTP on the ALB’s DNS nameHTTPS on your domainHTTPS on your domain
Backend scaling1 task, no autoscaling1-task floor and autoscaling2-task floor and autoscaling
RDSSingle-AZSingle-AZ and RDS ProxyMulti-AZ, RDS Proxy and secret rotation
RDS deletion protectionOffOnOn
RDS backup retention7 days7 days7 days
CognitoEnd-user client and admin groups, MFA offEnd-user client and admin groups, MFA optional (software token), deletion protectionAs production
AgentsThe assistant (code configuration)The assistantThe assistant
Agent network modePUBLICVPCVPC
AgentCore Memory retention30 days90 days90 days
Default seed profiledemofoundationfoundation
Rotation Opnoneloom-rotate-productionloom-rotate-production-ha

The same on every tier: the six components, cross-stack wiring, the cost tags, the naming scheme, and secret encryption with no opt-out. Seams are independent of the tier (Adoption seams).

TargetChosen byWhat runsVerify phase
FlociAWS_ENDPOINT_URL set, such as http://localhost:4566The stacks, including the agents wave on FlociSkipped: Floci doesn’t serve the ALB to ECS path
Real AWSAWS_ENDPOINT_URL unsetEverythingSteady state and the health check

just local-up is a third path: it deploys the light tier on Floci and runs the app from a generated compose file with a proxy in place of the ALB (Run Loom locally).

All three tiers keep 7 days of automated backups with point-in-time recovery, so for data loss the recovery point is about 5 minutes anywhere in that window. They differ in how an infrastructure failure is handled.

TierAZ or instance failureData loss (corruption, bad delete)Region loss
lightloom-restore and cutoverpoint-in-time or snapshot loom-restoresurvivable only with a loom-backup DR copy
productionloom-restore and cutoverpoint-in-time or snapshot loom-restoresurvivable only with a DR copy
production-haMulti-AZ failover, about 1 to 2 minutes, no data losspoint-in-time or snapshot loom-restoresurvivable only with a DR copy

Recovery time outside a Multi-AZ failover is how long the gated loom-restore takes to restore, wait for approval and cut over. Data older than the 7-day window survives only in a manual snapshot, which loom-backup takes. Back up and restore has the steps.