ReferenceAsk
Remy Reference/Architecture/Architecture/Data Inventory
Appendix

Data Inventory

Every store: contents, isolation, encryption, recovery, residency

Every place data rests, what it holds, the unit of isolation, how it is protected, how it is recovered, and where it lives.

StoreContentsIsolation unitEncryptionDurability and recoveryResidencyChapter
Aurora PostgreSQLWorkspaces, users, apps, releases, manifests, catalog of app databases, encrypted secrets, API key hashes, sessions, cron and webhook definitions, request logs, metrics, billing, audit indexes, settingsRow-level by workspace and appCustomer-managed KMS key at rest; TLS requiredContinuous backups with point-in-time recovery; daily cross-region copy, 28 days; weekly restore testCompute region09
App databases (SQLite on S3)Every app's tables and rowsOne file per app database, per releaseServer-side encryption at rest; HTTPS onlyFlushed within seconds of a write; versioned bucket; DDL on a clone at deploy; prior release's file retainedWorkspace's region, with a compute-region working copy while hot10
S3 private bucketsApp databases, git repository tars and file mirrors, build artifacts, commit diffs, private app files, workspace snapshots, presentation artifactsOwner-rooted keys per app and storeServer-side encryption; public access blocked; HTTPS onlyVersioned; newest prior versions retained; access-loggedOne bucket per region11
S3 public bucketsPublic app files, user uploads, web assetsOwner-rooted keysServer-side encryption; HTTPS onlyVersioned; prior versions expire after seven daysOne bucket per region11
S3 audit bucketAudit events as NDJSON, partitioned by workspace and datePrefix per workspaceServer-side encryption; public access blocked; HTTPS onlyObject Lock, COMPLIANCE, one year; no deletion possible before expiryCompute region19
S3 access-log bucketsServer access logs for private bucketsPer source bucketServer-side encryptionLifecycle expiryOne per region11
ValkeyCache, pub/sub, ownership records, pool registry, locks and leases, interface sessions, hook tokens, live audit tail, vendor limitsKey namespacesIn-VPCRebuildable from Postgres, S3 and pods; Multi-AZ with failoverCompute region13
SQSWork messagesQueue per work typeIn transitDead-letter queues on ingest paths; visibility heartbeatsCompute region13
Shared QdrantVector and lexical indexes for data sourcesTenant per data sourceLocal NVMe in the clusterA cache of S3 artifacts; rebuilt on cold read or evictionCompute region33
Dedicated QdrantOne customer's indexesInstance per resource, namespace and NetworkPolicyLocal disk in the clusterA cache of the customer's S3 artifacts; rebuilt when a pod comes up emptyCompute region33
Git repositoriesEach app's source historyRepository per appS3 server-side encryption for the durable tar; NVMe scratch on the podS3 tar is the source of truth; local copies are cachesWorkspace's region, with a compute-region working copy while hot14
Secrets ManagerWorkspaces' own vendor keysOne bundle per workspace, versionedAWS-managed encryptionVersioned; cache keyed by versionCompute region18
Parameter StoreThe platform's operator credentialsPer parameterAWS-managed encryptionSynced to the cluster on a short intervalCompute region18
App secrets (in Postgres)Builders' secrets, dev and prod valuesPer app and keyEnvelope-encrypted with KMS before storageWith the platform databaseCompute region18
CognitoPlatform user identitiesUser poolAWS-managedDeletion protection onCompute region17
Training bases and adaptersBase model weights; per-method LoRA adapters and evaluation reportsAdapters in the app's own file storeS3 server-side encryptionVersioned with the file storeWorkspace's region for adapters34
Sandbox local diskHandler files, extracted dependencies, a dev box's working copyPer microVMEphemeralDev boxes snapshot their home directory to S3 on shutdownCompute region while running06, 08
CloudWatch LogsDatabase audit log, Firehose delivery logs, Container Insights metricsLog groupAWS-managedRetention per groupCompute region25
CloudTrail bucketControl-plane API historyAccountServer-side encryption; log-file validationRetainedCompute region25
New Relic, SentryPerformance telemetry and error reportsAccountVendorVendor retention; both declared in the counterparty registry with their agreementsVendor25