Skip to content
loomster

Tiers and instances

llms.txt lists every page for a coding agent.

Every physical name and cost tag comes from one call, loomNaming(params, component), keyed {project}-{env}-{instance}-{component}-{resource} (Naming and tagging). Two of those inputs are the settings you choose for a deployment, and they don’t depend on each other.

Tier

Section titled “Tier”

LOOM_TIER is light, production or production-ha. It sizes one Loom:

  • light builds its own VPC with public subnets, serves HTTP on the ALB’s own DNS name, and runs single-AZ RDS. It needs nothing from you but an account.
  • production needs your VPC with private subnets, and adds HTTPS on a custom domain, PrivateLink, an RDS Proxy, backend autoscaling, Cognito MFA and deletion protection.
  • production-ha adds Multi-AZ RDS, Secrets Manager rotation of the database credential and a two-task backend floor.

Every tier runs the same source. No tier has its own files; each difference is a value resolved from the tier. The full table is in Tiers and targets.

Instance

Section titled “Instance”

LOOM_INSTANCE names which Loom this is. It’s in every physical name and every CloudFormation stack name (loom-prod-a-loom-db), so two instances in one account and region never collide, and loom-teardown only ever deletes its own instance’s stacks. LOOM_ENV works the same way one level up.

Topology

Section titled “Topology”

Loom has group-based multi-tenancy inside one deployment: Cognito groups give a user scopes, and the loom:group tag decides which resources they see. That isolation is logical. One Loom is one control plane and one database.

So there are two topologies:

  • Single-boundary: one Loom, many groups. Every team fits inside one account and compliance boundary and sees a scoped view.
  • Multi-boundary: many Looms, one instance per account, per prod and non-prod, or per compliance domain. A shared Cognito pool or external IdP can serve all of them, with groups and scopes defined once (Share a Cognito pool).

The rule of thumb is groups inside a boundary, and a new instance per boundary. If two teams must never see each other’s agents, give them separate instances. If they should see a scoped view of one platform, give them groups in one Loom.

Target

Section titled “Target”

Where a tier deploys is a third, separate choice: Floci when AWS_ENDPOINT_URL is set, real AWS when it isn’t. Any tier can go to either.