Tutorial
From a browsable local Loom to the light tier on real AWS to production. One codebase, tier and target as parameters. Start here.
loomster
Typed, tiered infrastructure-as-code for awslabs/loom on chant. Six components, three tiers, generated CI, and a naming scheme that lets many Loom instances coexist in one or many AWS accounts without collision.
What this is
Section titled “What this is”Typed infrastructure-as-code for running Loom,
built to the same bar as a production stack. Loom’s own deploy today is a manual,
multi-step SAM process behind a DEPLOYMENT.md. chant types it, lints it, dedupes
the cross-stack glue, orders it, tiers it, and generates the pipeline.
Six components, deployed in dependency order (shared-foundation, loom-cognito,
loom-db, loom-frontend, loom-backend, loom-agents), pinned to Loom
v1.6.0.
Run it locally
Section titled “Run it locally”just local-up # browsable, authenticated Loom on your laptop (Docker, no AWS account)Floci (a local AWS emulator) provides RDS, Cognito, S3, and ECR; the app runs
from a chant-generated compose, browsable at localhost:8080. The web app runs
for real. Only agents need real AWS. See
Run Loom on your laptop.
Where it runs today: the full stack runs on Floci, the light tier is deployed
end to end to a real AWS account, and production is validated end to end on a
real account — 7/7 stacks CREATE_COMPLETE, including the agents wave (the Strands
assistant’s code-config Runtime reaches READY on Bedrock AgentCore, app served over
HTTPS). production-ha is validated end to end on a real account too — 7/7 stacks, adding
Multi-AZ RDS + a live credential rotation, and it ran as a second instance alongside
production (multi-deployment). The
Overview has the full picture.
Same discipline, one layer down
Section titled “Same discipline, one layer down”Loom’s own launch post states its model plainly: no code is generated at runtime and deployed into any environment. Only configuration changes, and the control plane manages that configuration. That is chant’s pitch too, one layer beneath:
- Loom scans the agent code once, ahead of any deployment. chant type-checks and lints a stack at author time, before anything synthesizes.
- Loom never generates agent code at runtime. chant never generates
infrastructure code at deploy time.
chant buildemits a CloudFormation template once, and every environment applies that same template. - Loom redeploys by changing config, not by hand-editing a running agent. chant redeploys by changing tier, instance, or an adoption seam’s mode.
Tier and topology, two independent dials
Section titled “Tier and topology, two independent dials”Every physical name and cost tag traces back to one call,
loomNaming(params, component), keyed
{project}-{env}-{instance}-{component}-{resource}. Two axes, independent of each
other:
- Tier (
light/production/production-ha) sizes one Loom. HTTP vs. HTTPS + PrivateLink, single-AZ RDS vs. Multi-AZ + RDS Proxy + rotation, one agent vs. two. - Instance is the tenant/boundary segment, the topology axis. Two instance values are two collision-free Looms in the same account and region.
Rule of thumb: groups inside a boundary, a new Loom per boundary. See the Tutorial’s Org topology section.
Run Loom on your laptop
just local-up runs a browsable, authenticated Loom against Floci, no AWS account. What’s real, what’s a dev shortcut, and how it’s wired. The local guide.
Adoption
Every composite exposes a provision | reference-existing | omit choice where meaningful. Bring your own VPC, KMS key, ACM cert, Route53 zone, ECR repos, agent IAM role, Postgres endpoint, or Cognito pool. The full matrix.
Naming & Tagging
One shared parameter source for every physical resource name and cost-allocation tag. Segment order, per-service limits, uniqueness strategy, and lint enforcement. Read the convention.