Supervised deployments for AWS

for builders shipping to AWS with coding agents — no devops team required

Delegate your infra to your agent. Reliably.

Your agent deploys. Waypointer verifies — before apply and after. Independent, read-only, no migration.

Waypointer scans your AWS, you confirm the map once — and every plan your agent writes is checked against it.

Read-only IAM role · nothing to migrate · five minutes to connect

THE CATCH waypointer check_plan
task: “clean up the old database replica”
agent wrote a destroy plan · asks waypointer to check it… ✗ PLAN CHECK — do not apply
aws_db_instance  db-prod-01 — the PRIMARY database
  group: database / web-app / prod · you likely meant:
  db-prod-replica-02 agent stopped before apply · valid Terraform, wrong resource
A CLEAN DEPLOY waypointer check_plan
task: “add a dead-letter queue to the billing worker” ✓ PLAN CHECK — safe
+ 2 resources → group: queue / billing-worker / prod
  untouched: database group (20), shared ALB, all other apps
✓ post-apply scan — reality matches the plan
logged → history: billing-worker / prod · note saved

The problem

Your agent can do everything right — to the wrong resource.

The code it writes is just plumbing — you never read it. Waypointer turns every change into a plain verdict: what, where, which app, what’s untouched.

STOPPED BEFORE APPLY

Agent about to delete the wrong database?

The plan is checked against what is actually running — not against a state file that may already be wrong. The verdict is in plain words: app, environment, what’s affected — no Terraform to read.

Valid Terraform aimed at the primary instance stops here.

CAUGHT IN MINUTES

Did the apply actually do what it said?

“Apply complete” doesn’t mean reality matches — timeouts, half-applied changes and silent defaults happen. Waypointer reads AWS after every apply and shows what actually landed.

Two resources promised, one created — you hear it from us, not from the bill.

NEVER AGAIN

Managing several projects? Re-explaining your infra every session?

You don’t start from zero — even a month later. The map is waiting: every resource, its app, its environment, and the history of what changed while you were away.

“What did we do to billing last week?” — answered before you finish typing it.

How Waypointer knows what your agent is touching

Repo, application, environment, cloud account — joined once and kept joined. Your agent works across the top; Waypointer runs underneath and checks every link it touches.

your agent repo app env cloud Waypointer

“What if I already have a cloud full of resources — and nothing linking them?”

That’s the default, and that’s what the Sherlock skill is for: your agent deduces which resource belongs to which app and environment — and shows you the evidence. You just confirm.

Security

One read-only IAM role

Cross-account role with a least-privilege policy: describe/list only, no data-plane reads, no PassRole, scoped by ExternalID.

Physically unable to change anything

No create, modify, or delete permissions exist on the role. Blocking happens in your agent — Waypointer only returns the verdict.

Your keys stay yours

Deploys run with your agent’s credentials on your machine. Waypointer never holds write access to your AWS.

“If Waypointer is read-only — how does it stop anything?”

It doesn’t need write access to stop a mistake: your agent checks with Waypointer before it applies, and a “do not apply” verdict is the stop. Waypointer never writes to AWS — deploys run with your agent’s credentials, on your machine.

Delegate your infra to your agent. Reliably.

ships with your invite
$ npx @waypointer/cli connect

Read-only IAM role · nothing to migrate · private beta