Skip to content

The cross-harness, self-improving memory layer for AI agents. Join our community: https://discord.com/invite/zdNChS2fBu

2.1 KB
Raw Blame

Terraform for Beacon object storage forwarding

These modules create the cloud side of Beacon's object storage forwarding: the bucket your Macs upload to, and the narrowly scoped identity their forwarder uses. You run them in your own cloud account. Beacon itself never runs Terraform and never needs your cloud credentials.

Module Creates Guide
aws-s3 A new S3 bucket and a write-only IAM user AWS S3 with Terraform
gcs A new Cloud Storage bucket and a create-only service account Google Cloud Storage with Terraform
aws-s3-asymptote-reader Read-only access for Asymptote to an aws-s3 bucket, with new-object SQS queues Asymptote with your own S3 bucket
gcs-asymptote-reader Read access for Asymptote to a gcs bucket, with new-object Pub/Sub subscriptions Asymptote with your own Cloud Storage bucket

The reader modules take the base module's bucket output, so the bucket is always one the base module created.

Use a module from your own configuration and pin it to a Beacon release:

module "beacon_s3" {
  source = "github.com/asymptote-labs/agent-beacon//terraform/aws-s3?ref=<release tag>"

  bucket_name = "acme-beacon-telemetry"
  prefix      = "beacon-prod"
}

No module creates keys unless you ask it to (create_writer_access_key, create_writer_key, create_access_key), so by default no secret is written to Terraform state. The guides show how to create the keys with the AWS CLI or gcloud instead.

Development

The tests mock the cloud providers, so they need no account:

cd terraform/aws-s3   # or any other module directory
terraform init -backend=false
terraform test

CI runs terraform fmt -check, validate and test for every module, and validate for each example.