diff --git a/servers/gcp/artifact-registry.tf b/servers/gcp/artifact-registry.tf
new file mode 100644
index 0000000..2791ccd
--- /dev/null
+++ b/servers/gcp/artifact-registry.tf
@@ -0,0 +1,96 @@
+# ── Artifact Registry ─────────────────────────────────────────────────────────
+# One repository per service. Images are pushed here from Legion CI runners
+# (the only compiled artifacts that cross the boundary to GCP).
+# Source code and intelligence stay on Legion — only the built container
+# crosses out.
+
+resource "google_artifact_registry_repository" "marketing" {
+  location      = "us-central1"
+  repository_id = "neuron-marketing"
+  description   = "Marketing site (Next.js) Docker images"
+  format        = "DOCKER"
+  project       = var.project_id
+
+  cleanup_policies {
+    id     = "keep-last-10"
+    action = "KEEP"
+    most_recent_versions {
+      keep_count = 10
+    }
+  }
+}
+
+resource "google_artifact_registry_repository" "accounts" {
+  location      = "us-central1"
+  repository_id = "neuron-accounts"
+  description   = "Accounts service (Go) Docker images"
+  format        = "DOCKER"
+  project       = var.project_id
+
+  cleanup_policies {
+    id     = "keep-last-10"
+    action = "KEEP"
+    most_recent_versions {
+      keep_count = 10
+    }
+  }
+}
+
+resource "google_artifact_registry_repository" "api" {
+  location      = "us-central1"
+  repository_id = "neuron-api"
+  description   = "REST API (neuron-rest, Go) Docker images"
+  format        = "DOCKER"
+  project       = var.project_id
+
+  cleanup_policies {
+    id     = "keep-last-10"
+    action = "KEEP"
+    most_recent_versions {
+      keep_count = 10
+    }
+  }
+}
+
+# ── IAM: CI runner pushes images ──────────────────────────────────────────────
+# The CI service account (from Legion's Gitea runner) needs Artifact Registry
+# writer access so it can push compiled images after each build.
+
+resource "google_service_account" "ci_pusher" {
+  account_id   = "neuron-ci-pusher"
+  display_name = "Neuron CI Image Pusher"
+  description  = "Legion CI runner uses this SA to push compiled images to Artifact Registry"
+  project      = var.project_id
+}
+
+resource "google_artifact_registry_repository_iam_member" "ci_marketing" {
+  project    = var.project_id
+  location   = "us-central1"
+  repository = google_artifact_registry_repository.marketing.name
+  role       = "roles/artifactregistry.writer"
+  member     = "serviceAccount:${google_service_account.ci_pusher.email}"
+}
+
+resource "google_artifact_registry_repository_iam_member" "ci_accounts" {
+  project    = var.project_id
+  location   = "us-central1"
+  repository = google_artifact_registry_repository.accounts.name
+  role       = "roles/artifactregistry.writer"
+  member     = "serviceAccount:${google_service_account.ci_pusher.email}"
+}
+
+resource "google_artifact_registry_repository_iam_member" "ci_api" {
+  project    = var.project_id
+  location   = "us-central1"
+  repository = google_artifact_registry_repository.api.name
+  role       = "roles/artifactregistry.writer"
+  member     = "serviceAccount:${google_service_account.ci_pusher.email}"
+}
+
+# ── CI pusher JSON key (download once, store in Vault/Secret Manager) ─────────
+# After `terraform apply`, create a key and save it to GCP Secret Manager:
+#   gcloud iam service-accounts keys create /tmp/ci-pusher.json \
+#     --iam-account=$(terraform output -raw ci_pusher_email)
+#   gcloud secrets create ci-gcp-sa-key --data-file=/tmp/ci-pusher.json
+#   rm /tmp/ci-pusher.json
+# Then set it as a repo secret in Gitea: GCP_SA_KEY
diff --git a/servers/gcp/bootstrap.sh b/servers/gcp/bootstrap.sh
new file mode 100755
index 0000000..297a7ae
--- /dev/null
+++ b/servers/gcp/bootstrap.sh
@@ -0,0 +1,97 @@
+#!/usr/bin/env bash
+# ── GCP Bootstrap — run once after terraform apply ────────────────────────────
+# Sets up secrets and credentials that Terraform can't create automatically.
+#
+# Prerequisites:
+#   - GOOGLE_APPLICATION_CREDENTIALS or `gcloud auth application-default login`
+#   - GCP project access
+#   - Vault running (for reading existing secrets)
+#   - terraform apply has run successfully (creates SA, secrets)
+#
+# Usage:
+#   cd servers/gcp
+#   terraform apply
+#   bash bootstrap.sh
+
+set -euo pipefail
+
+PROJECT="neuron-785695"
+VAULT_ADDR="${VAULT_ADDR:-https://vault.neuralplatform.ai}"
+
+echo "==> Bootstrapping GCP secrets and CI credentials"
+
+# ── 1. Populate Stripe secrets in GCP Secret Manager ─────────────────────────
+echo "--> Pulling Stripe secrets from Vault..."
+MARKETING_SECRETS=$(curl -sf \
+  -H "X-Vault-Token: ${VAULT_TOKEN}" \
+  "${VAULT_ADDR}/v1/secret/data/neuron-technologies/marketing")
+
+populate_secret() {
+  local secret_id="$1"
+  local vault_key="$2"
+  local value
+  value=$(echo "${MARKETING_SECRETS}" | python3 -c "
+import json,sys; d=json.load(sys.stdin)['data']['data']; print(d.get('${vault_key}',''))
+")
+  if [ -z "${value}" ]; then
+    echo "  WARN: vault key '${vault_key}' not found — skipping ${secret_id}"
+    return
+  fi
+  # Check if secret already has a version
+  if gcloud secrets versions list "${secret_id}" --project="${PROJECT}" \
+       --format="value(name)" 2>/dev/null | grep -q .; then
+    echo "  SKIP: ${secret_id} already has a version"
+  else
+    echo "${value}" | gcloud secrets versions add "${secret_id}" \
+      --project="${PROJECT}" \
+      --data-file=-
+    echo "  OK: ${secret_id}"
+  fi
+}
+
+populate_secret "stripe-secret-key"        "stripe_secret_key"
+populate_secret "stripe-webhook-secret"    "stripe_webhook_secret"
+populate_secret "stripe-price-professional" "stripe_price_professional"
+populate_secret "stripe-price-founding"    "stripe_price_founding"
+
+# ── 2. License admin token ────────────────────────────────────────────────────
+echo "--> Populating license-admin-token..."
+LICENSE_TOKEN=$(curl -sf \
+  -H "X-Vault-Token: ${VAULT_TOKEN}" \
+  "${VAULT_ADDR}/v1/secret/data/neuron-technologies/marketing" \
+  | python3 -c "import json,sys; print(json.load(sys.stdin)['data']['data'].get('license_admin_token',''))")
+if [ -n "${LICENSE_TOKEN}" ]; then
+  echo "${LICENSE_TOKEN}" | gcloud secrets versions add "license-admin-token" \
+    --project="${PROJECT}" --data-file=- 2>/dev/null || echo "  SKIP: already exists"
+  echo "  OK: license-admin-token"
+fi
+
+# ── 3. Create CI service account JSON key ────────────────────────────────────
+CI_SA_EMAIL=$(terraform output -raw ci_pusher_email)
+echo "--> Creating CI SA key for: ${CI_SA_EMAIL}"
+KEY_FILE="/tmp/neuron-ci-pusher.json"
+gcloud iam service-accounts keys create "${KEY_FILE}" \
+  --iam-account="${CI_SA_EMAIL}" \
+  --project="${PROJECT}"
+echo "  SA key written to: ${KEY_FILE}"
+echo ""
+echo "  !! Add this as a Gitea secret GCP_SA_KEY in the neuron-technologies/neuron repo:"
+echo "     tea secret set --repo neuron-technologies/neuron GCP_SA_KEY < ${KEY_FILE}"
+echo "  Then delete the key file:"
+echo "     rm ${KEY_FILE}"
+echo ""
+
+# ── 4. Summary ────────────────────────────────────────────────────────────────
+echo "==> Bootstrap complete"
+echo ""
+echo "Next steps:"
+echo "  1. Set Gitea secret: tea secret set --repo neuron-technologies/neuron GCP_SA_KEY < /tmp/neuron-ci-pusher.json"
+echo "  2. Delete key file: rm /tmp/neuron-ci-pusher.json"
+echo "  3. Point DNS at: $(terraform output -raw prod_lb_ip)"
+echo "     A records (proxied=true in Cloudflare):"
+echo "       neurontechnologies.ai → prod_lb_ip"
+echo "       www.neurontechnologies.ai → prod_lb_ip"
+echo "       api.neurontechnologies.ai → prod_lb_ip"
+echo "       accounts.neurontechnologies.ai → prod_lb_ip"
+echo "  4. Run a marketing CI build on main to push first image to GCP AR"
+echo "  5. Check SSL cert provisioning: gcloud compute ssl-certificates list --project=${PROJECT}"
diff --git a/servers/gcp/cloud-run-accounts.tf b/servers/gcp/cloud-run-accounts.tf
new file mode 100644
index 0000000..257a8f7
--- /dev/null
+++ b/servers/gcp/cloud-run-accounts.tf
@@ -0,0 +1,443 @@
+# ── Accounts Service — Cloud Run ──────────────────────────────────────────────
+# Handles auth, billing, subscriptions, marketplace.
+# Connects to Cloud SQL via built-in Auth Proxy (Unix socket volume mount).
+# All three regions share the same Cloud SQL instance (us-central1).
+
+locals {
+  accounts_labels = {
+    "managed-by" = "terraform"
+    "service"    = "neuron-accounts"
+  }
+  sql_instance = google_sql_database_instance.main.connection_name
+}
+
+# ── Prod — us-central1 ────────────────────────────────────────────────────────
+
+resource "google_cloud_run_v2_service" "accounts_us" {
+  name     = "accounts-prod-us"
+  location = "us-central1"
+  project  = var.project_id
+  ingress  = "INGRESS_TRAFFIC_ALL"
+  labels   = local.accounts_labels
+
+  template {
+    service_account = google_service_account.accounts.email
+
+    scaling {
+      min_instance_count = 1
+      max_instance_count = 50
+    }
+
+    containers {
+      image = local.accounts_image
+
+      resources {
+        limits = {
+          cpu    = "1"
+          memory = "512Mi"
+        }
+        cpu_idle = true
+      }
+
+      env {
+        name  = "ENV"
+        value = "production"
+      }
+      env {
+        name  = "PORT"
+        value = "8080"
+      }
+
+      env {
+        name = "ACCOUNTS_DATABASE_URL"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.accounts_database_url.secret_id
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "JWT_SECRET"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.jwt_secret.secret_id
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_SECRET_KEY"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-secret-key"
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_WEBHOOK_SECRET"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-webhook-secret"
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_PRICE_PROFESSIONAL"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-price-professional"
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_PRICE_FOUNDING"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-price-founding"
+            version = "latest"
+          }
+        }
+      }
+
+      ports {
+        container_port = 8080
+        name           = "http1"
+      }
+
+      startup_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        initial_delay_seconds = 2
+        timeout_seconds       = 5
+        period_seconds        = 5
+        failure_threshold     = 10
+      }
+
+      liveness_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        timeout_seconds   = 5
+        period_seconds    = 30
+        failure_threshold = 3
+      }
+
+      volume_mounts {
+        name       = "cloudsql"
+        mount_path = "/cloudsql"
+      }
+    }
+
+    volumes {
+      name = "cloudsql"
+      cloud_sql_instance {
+        instances = [local.sql_instance]
+      }
+    }
+  }
+
+  traffic {
+    type    = "TRAFFIC_TARGET_ALLOCATION_TYPE_LATEST"
+    percent = 100
+  }
+
+  depends_on = [
+    google_project_iam_member.accounts_secret_accessor,
+    google_project_iam_member.accounts_sql_client,
+    google_secret_manager_secret_version.accounts_database_url,
+    google_secret_manager_secret_version.jwt_secret,
+  ]
+}
+
+# ── Prod — europe-west1 ───────────────────────────────────────────────────────
+
+resource "google_cloud_run_v2_service" "accounts_eu" {
+  name     = "accounts-prod-eu"
+  location = "europe-west1"
+  project  = var.project_id
+  ingress  = "INGRESS_TRAFFIC_ALL"
+  labels   = local.accounts_labels
+
+  template {
+    service_account = google_service_account.accounts.email
+
+    scaling {
+      min_instance_count = 1
+      max_instance_count = 50
+    }
+
+    containers {
+      image = local.accounts_image
+
+      resources {
+        limits = {
+          cpu    = "1"
+          memory = "512Mi"
+        }
+        cpu_idle = true
+      }
+
+      env {
+        name  = "ENV"
+        value = "production"
+      }
+      env {
+        name  = "PORT"
+        value = "8080"
+      }
+
+      env {
+        name = "ACCOUNTS_DATABASE_URL"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.accounts_database_url.secret_id
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "JWT_SECRET"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.jwt_secret.secret_id
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_SECRET_KEY"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-secret-key"
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_WEBHOOK_SECRET"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-webhook-secret"
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_PRICE_PROFESSIONAL"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-price-professional"
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_PRICE_FOUNDING"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-price-founding"
+            version = "latest"
+          }
+        }
+      }
+
+      ports {
+        container_port = 8080
+        name           = "http1"
+      }
+
+      startup_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        initial_delay_seconds = 2
+        timeout_seconds       = 5
+        period_seconds        = 5
+        failure_threshold     = 10
+      }
+
+      liveness_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        timeout_seconds   = 5
+        period_seconds    = 30
+        failure_threshold = 3
+      }
+
+      volume_mounts {
+        name       = "cloudsql"
+        mount_path = "/cloudsql"
+      }
+    }
+
+    volumes {
+      name = "cloudsql"
+      cloud_sql_instance {
+        instances = [local.sql_instance]
+      }
+    }
+  }
+
+  traffic {
+    type    = "TRAFFIC_TARGET_ALLOCATION_TYPE_LATEST"
+    percent = 100
+  }
+
+  depends_on = [
+    google_project_iam_member.accounts_secret_accessor,
+    google_project_iam_member.accounts_sql_client,
+  ]
+}
+
+# ── Prod — asia-northeast1 ────────────────────────────────────────────────────
+
+resource "google_cloud_run_v2_service" "accounts_apac" {
+  name     = "accounts-prod-apac"
+  location = "asia-northeast1"
+  project  = var.project_id
+  ingress  = "INGRESS_TRAFFIC_ALL"
+  labels   = local.accounts_labels
+
+  template {
+    service_account = google_service_account.accounts.email
+
+    scaling {
+      min_instance_count = 1
+      max_instance_count = 50
+    }
+
+    containers {
+      image = local.accounts_image
+
+      resources {
+        limits = {
+          cpu    = "1"
+          memory = "512Mi"
+        }
+        cpu_idle = true
+      }
+
+      env {
+        name  = "ENV"
+        value = "production"
+      }
+      env {
+        name  = "PORT"
+        value = "8080"
+      }
+
+      env {
+        name = "ACCOUNTS_DATABASE_URL"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.accounts_database_url.secret_id
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "JWT_SECRET"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.jwt_secret.secret_id
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_SECRET_KEY"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-secret-key"
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_WEBHOOK_SECRET"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-webhook-secret"
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_PRICE_PROFESSIONAL"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-price-professional"
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "STRIPE_PRICE_FOUNDING"
+        value_source {
+          secret_key_ref {
+            secret  = "stripe-price-founding"
+            version = "latest"
+          }
+        }
+      }
+
+      ports {
+        container_port = 8080
+        name           = "http1"
+      }
+
+      startup_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        initial_delay_seconds = 2
+        timeout_seconds       = 5
+        period_seconds        = 5
+        failure_threshold     = 10
+      }
+
+      liveness_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        timeout_seconds   = 5
+        period_seconds    = 30
+        failure_threshold = 3
+      }
+
+      volume_mounts {
+        name       = "cloudsql"
+        mount_path = "/cloudsql"
+      }
+    }
+
+    volumes {
+      name = "cloudsql"
+      cloud_sql_instance {
+        instances = [local.sql_instance]
+      }
+    }
+  }
+
+  traffic {
+    type    = "TRAFFIC_TARGET_ALLOCATION_TYPE_LATEST"
+    percent = 100
+  }
+
+  depends_on = [
+    google_project_iam_member.accounts_secret_accessor,
+    google_project_iam_member.accounts_sql_client,
+  ]
+}
diff --git a/servers/gcp/cloud-run-api.tf b/servers/gcp/cloud-run-api.tf
new file mode 100644
index 0000000..d956916
--- /dev/null
+++ b/servers/gcp/cloud-run-api.tf
@@ -0,0 +1,308 @@
+# ── REST API (neuron-rest) — Cloud Run ────────────────────────────────────────
+# Stateless API layer. Validates JWTs from accounts service.
+# No database connection — reads user context from accounts service.
+# cpu_idle=false: always-hot, no cold-start latency for API calls.
+
+locals {
+  api_labels = {
+    "managed-by" = "terraform"
+    "service"    = "neuron-api"
+  }
+}
+
+# ── Prod — us-central1 ────────────────────────────────────────────────────────
+
+resource "google_cloud_run_v2_service" "api_us" {
+  name     = "api-prod-us"
+  location = "us-central1"
+  project  = var.project_id
+  ingress  = "INGRESS_TRAFFIC_ALL"
+  labels   = local.api_labels
+
+  template {
+    service_account = google_service_account.api.email
+
+    scaling {
+      min_instance_count = 1
+      max_instance_count = 100
+    }
+
+    containers {
+      image = local.api_image
+
+      resources {
+        limits = {
+          cpu    = "2"
+          memory = "1Gi"
+        }
+        cpu_idle = false
+      }
+
+      env {
+        name  = "ENV"
+        value = "production"
+      }
+      env {
+        name  = "PORT"
+        value = "8080"
+      }
+      env {
+        name  = "ACCOUNTS_SERVICE_URL"
+        value = "https://accounts.neurontechnologies.ai"
+      }
+
+      env {
+        name = "JWT_SECRET"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.jwt_secret.secret_id
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "LICENSE_ADMIN_TOKEN"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.license_admin_token.secret_id
+            version = "latest"
+          }
+        }
+      }
+
+      ports {
+        container_port = 8080
+        name           = "http1"
+      }
+
+      startup_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        initial_delay_seconds = 2
+        timeout_seconds       = 5
+        period_seconds        = 5
+        failure_threshold     = 6
+      }
+
+      liveness_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        timeout_seconds   = 5
+        period_seconds    = 30
+        failure_threshold = 3
+      }
+    }
+
+    max_instance_request_concurrency = 1000
+  }
+
+  traffic {
+    type    = "TRAFFIC_TARGET_ALLOCATION_TYPE_LATEST"
+    percent = 100
+  }
+
+  depends_on = [
+    google_project_iam_member.api_secret_accessor,
+    google_secret_manager_secret_version.jwt_secret,
+  ]
+}
+
+# ── Prod — europe-west1 ───────────────────────────────────────────────────────
+
+resource "google_cloud_run_v2_service" "api_eu" {
+  name     = "api-prod-eu"
+  location = "europe-west1"
+  project  = var.project_id
+  ingress  = "INGRESS_TRAFFIC_ALL"
+  labels   = local.api_labels
+
+  template {
+    service_account = google_service_account.api.email
+
+    scaling {
+      min_instance_count = 1
+      max_instance_count = 100
+    }
+
+    containers {
+      image = local.api_image
+
+      resources {
+        limits = {
+          cpu    = "2"
+          memory = "1Gi"
+        }
+        cpu_idle = false
+      }
+
+      env {
+        name  = "ENV"
+        value = "production"
+      }
+      env {
+        name  = "PORT"
+        value = "8080"
+      }
+      env {
+        name  = "ACCOUNTS_SERVICE_URL"
+        value = "https://accounts.neurontechnologies.ai"
+      }
+
+      env {
+        name = "JWT_SECRET"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.jwt_secret.secret_id
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "LICENSE_ADMIN_TOKEN"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.license_admin_token.secret_id
+            version = "latest"
+          }
+        }
+      }
+
+      ports {
+        container_port = 8080
+        name           = "http1"
+      }
+
+      startup_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        initial_delay_seconds = 2
+        timeout_seconds       = 5
+        period_seconds        = 5
+        failure_threshold     = 6
+      }
+
+      liveness_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        timeout_seconds   = 5
+        period_seconds    = 30
+        failure_threshold = 3
+      }
+    }
+
+    max_instance_request_concurrency = 1000
+  }
+
+  traffic {
+    type    = "TRAFFIC_TARGET_ALLOCATION_TYPE_LATEST"
+    percent = 100
+  }
+
+  depends_on = [google_project_iam_member.api_secret_accessor]
+}
+
+# ── Prod — asia-northeast1 ────────────────────────────────────────────────────
+
+resource "google_cloud_run_v2_service" "api_apac" {
+  name     = "api-prod-apac"
+  location = "asia-northeast1"
+  project  = var.project_id
+  ingress  = "INGRESS_TRAFFIC_ALL"
+  labels   = local.api_labels
+
+  template {
+    service_account = google_service_account.api.email
+
+    scaling {
+      min_instance_count = 1
+      max_instance_count = 100
+    }
+
+    containers {
+      image = local.api_image
+
+      resources {
+        limits = {
+          cpu    = "2"
+          memory = "1Gi"
+        }
+        cpu_idle = false
+      }
+
+      env {
+        name  = "ENV"
+        value = "production"
+      }
+      env {
+        name  = "PORT"
+        value = "8080"
+      }
+      env {
+        name  = "ACCOUNTS_SERVICE_URL"
+        value = "https://accounts.neurontechnologies.ai"
+      }
+
+      env {
+        name = "JWT_SECRET"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.jwt_secret.secret_id
+            version = "latest"
+          }
+        }
+      }
+      env {
+        name = "LICENSE_ADMIN_TOKEN"
+        value_source {
+          secret_key_ref {
+            secret  = google_secret_manager_secret.license_admin_token.secret_id
+            version = "latest"
+          }
+        }
+      }
+
+      ports {
+        container_port = 8080
+        name           = "http1"
+      }
+
+      startup_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        initial_delay_seconds = 2
+        timeout_seconds       = 5
+        period_seconds        = 5
+        failure_threshold     = 6
+      }
+
+      liveness_probe {
+        http_get {
+          path = "/health"
+          port = 8080
+        }
+        timeout_seconds   = 5
+        period_seconds    = 30
+        failure_threshold = 3
+      }
+    }
+
+    max_instance_request_concurrency = 1000
+  }
+
+  traffic {
+    type    = "TRAFFIC_TARGET_ALLOCATION_TYPE_LATEST"
+    percent = 100
+  }
+
+  depends_on = [google_project_iam_member.api_secret_accessor]
+}
diff --git a/servers/gcp/cloud-sql.tf b/servers/gcp/cloud-sql.tf
new file mode 100644
index 0000000..f4c42eb
--- /dev/null
+++ b/servers/gcp/cloud-sql.tf
@@ -0,0 +1,150 @@
+# ── Cloud SQL — PostgreSQL 15 ─────────────────────────────────────────────────
+# Single instance in us-central1 for the accounts service.
+# Cloud Run services connect via the built-in Cloud SQL Auth Proxy
+# (no direct IP exposure, encrypted, IAM-authenticated).
+#
+# Sizing: db-g1-small (1 shared vCPU, 1.7 GB) for launch. Scale up to
+# db-n1-standard-2 once traffic warrants it — zero-downtime restart.
+
+resource "google_sql_database_instance" "main" {
+  name             = "neuron-prod-pg15"
+  database_version = "POSTGRES_15"
+  region           = "us-central1"
+  project          = var.project_id
+
+  # Prevent accidental deletion via terraform destroy
+  deletion_protection = true
+
+  settings {
+    tier              = "db-g1-small"
+    availability_type = "ZONAL"  # Upgrade to REGIONAL (HA) once > 1k users
+
+    backup_configuration {
+      enabled                        = true
+      point_in_time_recovery_enabled = true
+      start_time                     = "03:00"  # 3am UTC — before backup window on Legion
+      transaction_log_retention_days = 7
+      backup_retention_settings {
+        retained_backups = 14
+      }
+    }
+
+    maintenance_window {
+      day          = 7  # Sunday
+      hour         = 4  # 4am UTC
+      update_track = "stable"
+    }
+
+    ip_configuration {
+      # No public IP — Cloud Run uses the Auth Proxy via private service connect
+      # Flip ipv4_enabled=true + authorized_networks if you ever need direct access
+      # for migrations/seeding from a bastion.
+      ipv4_enabled    = true   # Required for Cloud Run Auth Proxy (until VPC SC configured)
+      ssl_mode        = "ENCRYPTED_ONLY"
+    }
+
+    database_flags {
+      name  = "max_connections"
+      value = "100"  # Cloud Run can burst many instances; cap connections early
+    }
+
+    database_flags {
+      name  = "log_min_duration_statement"
+      value = "500"  # Log queries taking > 500ms
+    }
+
+    insights_config {
+      query_insights_enabled  = true
+      query_string_length     = 1024
+      record_application_tags = true
+      record_client_address   = false
+    }
+  }
+}
+
+# ── Databases ─────────────────────────────────────────────────────────────────
+
+resource "google_sql_database" "accounts" {
+  name     = "accounts"
+  instance = google_sql_database_instance.main.name
+  project  = var.project_id
+}
+
+# ── Users ─────────────────────────────────────────────────────────────────────
+# Passwords are stored in Secret Manager. The accounts service reads them
+# at runtime via the ACCOUNTS_DATABASE_URL secret.
+# Cloud IAM DB users are the preferred path long-term (no password rotation);
+# using password auth here for compatibility with pgx/stdlib.
+
+resource "google_sql_user" "accounts" {
+  name     = "accounts"
+  instance = google_sql_database_instance.main.name
+  project  = var.project_id
+  password = random_password.accounts_db.result
+}
+
+resource "random_password" "accounts_db" {
+  length  = 32
+  special = true
+}
+
+# ── Secret Manager — Database URL ─────────────────────────────────────────────
+# Stored as a full DSN so the app just reads one env var.
+# Uses /cloudsql/ Unix socket path — Cloud Run Auth Proxy mounts it there.
+
+resource "google_secret_manager_secret" "accounts_database_url" {
+  secret_id = "accounts-database-url"
+  project   = var.project_id
+
+  replication {
+    auto {}
+  }
+}
+
+resource "google_secret_manager_secret_version" "accounts_database_url" {
+  secret = google_secret_manager_secret.accounts_database_url.id
+  secret_data = "host=/cloudsql/${google_sql_database_instance.main.connection_name} user=accounts password=${random_password.accounts_db.result} dbname=accounts sslmode=disable"
+}
+
+# ── Secret Manager — JWT signing key ─────────────────────────────────────────
+# Shared between accounts (issues tokens) and neuron-rest (validates tokens).
+
+resource "random_password" "jwt_secret" {
+  length  = 64
+  special = false  # URL-safe; used as HMAC key
+}
+
+resource "google_secret_manager_secret" "jwt_secret" {
+  secret_id = "jwt-signing-key"
+  project   = var.project_id
+
+  replication {
+    auto {}
+  }
+}
+
+resource "google_secret_manager_secret_version" "jwt_secret" {
+  secret      = google_secret_manager_secret.jwt_secret.id
+  secret_data = random_password.jwt_secret.result
+}
+
+# ── Secret Manager — License admin token ─────────────────────────────────────
+# Used by neuron-rest to authorize license validation calls.
+# Currently held in Vault under neuron-technologies/marketing.license_admin_token.
+# Create this secret manually from Vault:
+#   gcloud secrets create license-admin-token --data-file=<(vault kv get \
+#     -field=license_admin_token secret/neuron-technologies/marketing)
+
+resource "google_secret_manager_secret" "license_admin_token" {
+  secret_id = "license-admin-token"
+  project   = var.project_id
+
+  replication {
+    auto {}
+  }
+  # Version populated manually or via bootstrap script — not managed by Terraform
+  # to avoid exposing the value in state.
+  lifecycle {
+    ignore_changes = [id]
+  }
+}
diff --git a/servers/gcp/load-balancer-backends.tf b/servers/gcp/load-balancer-backends.tf
new file mode 100644
index 0000000..a7ab25e
--- /dev/null
+++ b/servers/gcp/load-balancer-backends.tf
@@ -0,0 +1,119 @@
+# ── Serverless NEGs — Accounts ────────────────────────────────────────────────
+
+resource "google_compute_region_network_endpoint_group" "accounts_us" {
+  name                  = "accounts-neg-us"
+  network_endpoint_type = "SERVERLESS"
+  region                = "us-central1"
+  project               = var.project_id
+  cloud_run { service = google_cloud_run_v2_service.accounts_us.name }
+}
+
+resource "google_compute_region_network_endpoint_group" "accounts_eu" {
+  name                  = "accounts-neg-eu"
+  network_endpoint_type = "SERVERLESS"
+  region                = "europe-west1"
+  project               = var.project_id
+  cloud_run { service = google_cloud_run_v2_service.accounts_eu.name }
+}
+
+resource "google_compute_region_network_endpoint_group" "accounts_apac" {
+  name                  = "accounts-neg-apac"
+  network_endpoint_type = "SERVERLESS"
+  region                = "asia-northeast1"
+  project               = var.project_id
+  cloud_run { service = google_cloud_run_v2_service.accounts_apac.name }
+}
+
+# ── Serverless NEGs — API ─────────────────────────────────────────────────────
+
+resource "google_compute_region_network_endpoint_group" "api_us" {
+  name                  = "api-neg-us"
+  network_endpoint_type = "SERVERLESS"
+  region                = "us-central1"
+  project               = var.project_id
+  cloud_run { service = google_cloud_run_v2_service.api_us.name }
+}
+
+resource "google_compute_region_network_endpoint_group" "api_eu" {
+  name                  = "api-neg-eu"
+  network_endpoint_type = "SERVERLESS"
+  region                = "europe-west1"
+  project               = var.project_id
+  cloud_run { service = google_cloud_run_v2_service.api_eu.name }
+}
+
+resource "google_compute_region_network_endpoint_group" "api_apac" {
+  name                  = "api-neg-apac"
+  network_endpoint_type = "SERVERLESS"
+  region                = "asia-northeast1"
+  project               = var.project_id
+  cloud_run { service = google_cloud_run_v2_service.api_apac.name }
+}
+
+# ── Backend Service — Accounts ────────────────────────────────────────────────
+
+resource "google_compute_backend_service" "accounts" {
+  name                  = "accounts-backend-prod"
+  project               = var.project_id
+  load_balancing_scheme = "EXTERNAL_MANAGED"
+  protocol              = "HTTPS"
+  timeout_sec           = 30
+
+  security_policy = google_compute_security_policy.marketing.self_link
+
+  # No CDN for accounts — responses are user-specific and must not be cached
+  enable_cdn = false
+
+  backend { group = google_compute_region_network_endpoint_group.accounts_us.self_link }
+  backend { group = google_compute_region_network_endpoint_group.accounts_eu.self_link }
+  backend { group = google_compute_region_network_endpoint_group.accounts_apac.self_link }
+
+  log_config {
+    enable      = true
+    sample_rate = 1.0
+  }
+}
+
+# ── Backend Service — API ─────────────────────────────────────────────────────
+
+resource "google_compute_backend_service" "api" {
+  name                  = "api-backend-prod"
+  project               = var.project_id
+  load_balancing_scheme = "EXTERNAL_MANAGED"
+  protocol              = "HTTPS"
+  timeout_sec           = 60  # API calls may be longer (AI inference routing)
+
+  security_policy = google_compute_security_policy.marketing.self_link
+
+  enable_cdn = false  # API responses are per-request; no CDN benefit
+
+  backend { group = google_compute_region_network_endpoint_group.api_us.self_link }
+  backend { group = google_compute_region_network_endpoint_group.api_eu.self_link }
+  backend { group = google_compute_region_network_endpoint_group.api_apac.self_link }
+
+  log_config {
+    enable      = true
+    sample_rate = 1.0
+  }
+}
+
+# ── SSL Certs — accounts and api subdomains ───────────────────────────────────
+# Added to the existing prod HTTPS proxy alongside the marketing cert.
+# DNS must point to the same prod global IP (marketing-ip-prod) for
+# provisioning to succeed.
+
+resource "google_compute_managed_ssl_certificate" "accounts" {
+  name    = "accounts-cert-prod"
+  project = var.project_id
+  managed {
+    domains = ["accounts.neurontechnologies.ai"]
+  }
+}
+
+resource "google_compute_managed_ssl_certificate" "api" {
+  name    = "api-cert-prod"
+  project = var.project_id
+  managed {
+    domains = ["api.neurontechnologies.ai"]
+  }
+}
diff --git a/servers/gcp/load-balancer.tf b/servers/gcp/load-balancer.tf
index 9a6c345..0aa21aa 100644
--- a/servers/gcp/load-balancer.tf
+++ b/servers/gcp/load-balancer.tf
@@ -223,21 +223,59 @@ resource "google_compute_managed_ssl_certificate" "stage" {
   }
 }
 
-# ── Prod URL Map ──────────────────────────────────────────────────────────────
+# ── Prod URL Map — host-based routing ────────────────────────────────────────
+# One global IP handles all three services via Host header routing.
+# - neurontechnologies.ai / www.  → marketing (default)
+# - api.neurontechnologies.ai     → API backend
+# - accounts.neurontechnologies.ai → accounts backend
 
 resource "google_compute_url_map" "prod" {
   name            = "marketing-urlmap-prod"
   project         = var.project_id
   default_service = google_compute_backend_service.prod.self_link
+
+  host_rule {
+    hosts        = ["neurontechnologies.ai", "www.neurontechnologies.ai"]
+    path_matcher = "marketing"
+  }
+
+  host_rule {
+    hosts        = ["api.neurontechnologies.ai"]
+    path_matcher = "api"
+  }
+
+  host_rule {
+    hosts        = ["accounts.neurontechnologies.ai"]
+    path_matcher = "accounts"
+  }
+
+  path_matcher {
+    name            = "marketing"
+    default_service = google_compute_backend_service.prod.self_link
+  }
+
+  path_matcher {
+    name            = "api"
+    default_service = google_compute_backend_service.api.self_link
+  }
+
+  path_matcher {
+    name            = "accounts"
+    default_service = google_compute_backend_service.accounts.self_link
+  }
 }
 
 # ── Prod HTTPS Target Proxy ───────────────────────────────────────────────────
 
 resource "google_compute_target_https_proxy" "prod" {
-  name             = "marketing-https-proxy-prod"
-  project          = var.project_id
-  url_map          = google_compute_url_map.prod.self_link
-  ssl_certificates = [google_compute_managed_ssl_certificate.prod.self_link]
+  name    = "marketing-https-proxy-prod"
+  project = var.project_id
+  url_map = google_compute_url_map.prod.self_link
+  ssl_certificates = [
+    google_compute_managed_ssl_certificate.prod.self_link,
+    google_compute_managed_ssl_certificate.accounts.self_link,
+    google_compute_managed_ssl_certificate.api.self_link,
+  ]
 }
 
 # ── Prod HTTP → HTTPS redirect ────────────────────────────────────────────────
diff --git a/servers/gcp/main.tf b/servers/gcp/main.tf
index 702fab2..47a109c 100644
--- a/servers/gcp/main.tf
+++ b/servers/gcp/main.tf
@@ -10,6 +10,10 @@ terraform {
       source  = "hashicorp/google-beta"
       version = "~> 5.0"
     }
+    random = {
+      source  = "hashicorp/random"
+      version = "~> 3.6"
+    }
   }
 
   backend "s3" {
diff --git a/servers/gcp/outputs.tf b/servers/gcp/outputs.tf
index 5613db7..1ab4933 100644
--- a/servers/gcp/outputs.tf
+++ b/servers/gcp/outputs.tf
@@ -1,44 +1,96 @@
 # ── Outputs ───────────────────────────────────────────────────────────────────
-# After `terraform apply`, use these IPs to update Cloudflare DNS A records.
+# After `terraform apply`, use prod_lb_ip for Cloudflare DNS A records.
 #
-# Prod:  neurontechnologies.ai + www.neurontechnologies.ai → prod_lb_ip
-# Stage: stage.neurontechnologies.ai                       → stage_lb_ip
+# All three services share the same global anycast IP (prod_lb_ip).
+# Set these A records in Cloudflare (proxied=true for CDN + DDoS):
+#   neurontechnologies.ai       → prod_lb_ip
+#   www.neurontechnologies.ai   → prod_lb_ip
+#   api.neurontechnologies.ai   → prod_lb_ip
+#   accounts.neurontechnologies.ai → prod_lb_ip
 #
-# In the Legion Terraform (dns-neurontechnologies.tf), replace the tunnel CNAMEs
-# for neurontechnologies.ai, www, and stage with A records pointing to these IPs.
-# Set proxied=true in Cloudflare to keep CDN/DDoS protection layered in front.
+# Stage: stage.neurontechnologies.ai → stage_lb_ip (separate IP)
 
 output "prod_lb_ip" {
-  description = "Global anycast IP for the production load balancer (neurontechnologies.ai + www)"
+  description = "Global anycast IP for all prod services (marketing, accounts, api)"
   value       = google_compute_global_address.prod.address
 }
 
 output "stage_lb_ip" {
-  description = "Global anycast IP for the staging load balancer (stage.neurontechnologies.ai)"
+  description = "Global anycast IP for the staging load balancer"
   value       = google_compute_global_address.stage.address
 }
 
 output "prod_ssl_cert_name" {
-  description = "Name of the Google-managed SSL cert for prod (check provisioning status in GCP console)"
+  description = "Marketing SSL cert (check provisioning status in GCP console)"
   value       = google_compute_managed_ssl_certificate.prod.name
 }
 
+output "accounts_ssl_cert_name" {
+  description = "Accounts SSL cert (check provisioning status in GCP console)"
+  value       = google_compute_managed_ssl_certificate.accounts.name
+}
+
+output "api_ssl_cert_name" {
+  description = "API SSL cert (check provisioning status in GCP console)"
+  value       = google_compute_managed_ssl_certificate.api.name
+}
+
 output "stage_ssl_cert_name" {
-  description = "Name of the Google-managed SSL cert for stage"
+  description = "Stage SSL cert"
   value       = google_compute_managed_ssl_certificate.stage.name
 }
 
 output "marketing_service_account_email" {
-  description = "Service account email for Cloud Run services"
+  description = "Marketing Cloud Run SA"
   value       = google_service_account.marketing.email
 }
 
+output "accounts_service_account_email" {
+  description = "Accounts Cloud Run SA"
+  value       = google_service_account.accounts.email
+}
+
+output "api_service_account_email" {
+  description = "API Cloud Run SA"
+  value       = google_service_account.api.email
+}
+
+output "ci_pusher_email" {
+  description = "CI SA email — use to create the JSON key for GCP_SA_KEY Gitea secret"
+  value       = google_service_account.ci_pusher.email
+}
+
+output "cloud_sql_connection_name" {
+  description = "Cloud SQL instance connection name — use in Cloud Run volume mounts"
+  value       = google_sql_database_instance.main.connection_name
+}
+
+output "cloud_sql_instance_ip" {
+  description = "Cloud SQL public IP (for bastion/migration access — disable when not needed)"
+  value       = google_sql_database_instance.main.public_ip_address
+}
+
 output "cloud_run_services" {
-  description = "Cloud Run service URLs per region"
+  description = "Cloud Run service URLs per service and region"
   value = {
-    prod_us   = google_cloud_run_v2_service.prod_us.uri
-    prod_eu   = google_cloud_run_v2_service.prod_eu.uri
-    prod_apac = google_cloud_run_v2_service.prod_apac.uri
-    stage     = google_cloud_run_v2_service.stage.uri
+    marketing_us   = google_cloud_run_v2_service.prod_us.uri
+    marketing_eu   = google_cloud_run_v2_service.prod_eu.uri
+    marketing_apac = google_cloud_run_v2_service.prod_apac.uri
+    marketing_stage = google_cloud_run_v2_service.stage.uri
+    accounts_us    = google_cloud_run_v2_service.accounts_us.uri
+    accounts_eu    = google_cloud_run_v2_service.accounts_eu.uri
+    accounts_apac  = google_cloud_run_v2_service.accounts_apac.uri
+    api_us         = google_cloud_run_v2_service.api_us.uri
+    api_eu         = google_cloud_run_v2_service.api_eu.uri
+    api_apac       = google_cloud_run_v2_service.api_apac.uri
+  }
+}
+
+output "artifact_registry_urls" {
+  description = "Docker image base URLs for each service"
+  value = {
+    marketing = "us-central1-docker.pkg.dev/${var.project_id}/neuron-marketing/marketing"
+    accounts  = "us-central1-docker.pkg.dev/${var.project_id}/neuron-accounts/accounts"
+    api       = "us-central1-docker.pkg.dev/${var.project_id}/neuron-api/api"
   }
 }
diff --git a/servers/gcp/service-account.tf b/servers/gcp/service-account.tf
index 70fcc90..1c0692e 100644
--- a/servers/gcp/service-account.tf
+++ b/servers/gcp/service-account.tf
@@ -1,13 +1,27 @@
-# ── Service Account ───────────────────────────────────────────────────────────
+# ── Service Accounts ──────────────────────────────────────────────────────────
 
 resource "google_service_account" "marketing" {
   account_id   = "neuron-marketing-sa"
   display_name = "Neuron Marketing Cloud Run SA"
   description  = "Service account for the neurontechnologies.ai marketing site on Cloud Run"
+  project      = var.project_id
 }
 
-# ── IAM — Secret Manager accessor ─────────────────────────────────────────────
-# Grant the SA access to read all marketing secrets from Secret Manager.
+resource "google_service_account" "accounts" {
+  account_id   = "neuron-accounts-sa"
+  display_name = "Neuron Accounts Cloud Run SA"
+  description  = "Service account for the accounts service (auth, billing, marketplace)"
+  project      = var.project_id
+}
+
+resource "google_service_account" "api" {
+  account_id   = "neuron-api-sa"
+  display_name = "Neuron REST API Cloud Run SA"
+  description  = "Service account for the neuron-rest API service"
+  project      = var.project_id
+}
+
+# ── IAM — Secret Manager ──────────────────────────────────────────────────────
 
 resource "google_project_iam_member" "marketing_secret_accessor" {
   project = var.project_id
@@ -15,9 +29,32 @@ resource "google_project_iam_member" "marketing_secret_accessor" {
   member  = "serviceAccount:${google_service_account.marketing.email}"
 }
 
-# ── IAM — Public (unauthenticated) invocation for prod services ───────────────
-# Cloud Run v2: allow-unauthenticated is set per-service via google_cloud_run_v2_service_iam_member.
+resource "google_project_iam_member" "accounts_secret_accessor" {
+  project = var.project_id
+  role    = "roles/secretmanager.secretAccessor"
+  member  = "serviceAccount:${google_service_account.accounts.email}"
+}
 
+resource "google_project_iam_member" "api_secret_accessor" {
+  project = var.project_id
+  role    = "roles/secretmanager.secretAccessor"
+  member  = "serviceAccount:${google_service_account.api.email}"
+}
+
+# ── IAM — Cloud SQL ───────────────────────────────────────────────────────────
+# Only accounts connects to Cloud SQL; API and marketing don't need DB access.
+
+resource "google_project_iam_member" "accounts_sql_client" {
+  project = var.project_id
+  role    = "roles/cloudsql.client"
+  member  = "serviceAccount:${google_service_account.accounts.email}"
+}
+
+# ── IAM — Public invocation (Cloud Run → LB) ──────────────────────────────────
+# allUsers invoker allows the global HTTP(S) load balancer's health checks
+# and traffic to reach the services without additional auth headers.
+
+# Marketing
 resource "google_cloud_run_v2_service_iam_member" "prod_us_public" {
   project  = var.project_id
   location = "us-central1"
@@ -48,7 +85,55 @@ resource "google_cloud_run_v2_service_iam_member" "stage_public" {
   name     = google_cloud_run_v2_service.stage.name
   role     = "roles/run.invoker"
   member   = "allUsers"
-
-  # Note: Cloudflare Access enforces authentication in front of stage.
-  # Cloud Run itself allows allUsers so the LB health checks pass.
+  # Note: Cloudflare Access enforces auth in front of stage.
+}
+
+# Accounts
+resource "google_cloud_run_v2_service_iam_member" "accounts_us_public" {
+  project  = var.project_id
+  location = "us-central1"
+  name     = google_cloud_run_v2_service.accounts_us.name
+  role     = "roles/run.invoker"
+  member   = "allUsers"
+}
+
+resource "google_cloud_run_v2_service_iam_member" "accounts_eu_public" {
+  project  = var.project_id
+  location = "europe-west1"
+  name     = google_cloud_run_v2_service.accounts_eu.name
+  role     = "roles/run.invoker"
+  member   = "allUsers"
+}
+
+resource "google_cloud_run_v2_service_iam_member" "accounts_apac_public" {
+  project  = var.project_id
+  location = "asia-northeast1"
+  name     = google_cloud_run_v2_service.accounts_apac.name
+  role     = "roles/run.invoker"
+  member   = "allUsers"
+}
+
+# API
+resource "google_cloud_run_v2_service_iam_member" "api_us_public" {
+  project  = var.project_id
+  location = "us-central1"
+  name     = google_cloud_run_v2_service.api_us.name
+  role     = "roles/run.invoker"
+  member   = "allUsers"
+}
+
+resource "google_cloud_run_v2_service_iam_member" "api_eu_public" {
+  project  = var.project_id
+  location = "europe-west1"
+  name     = google_cloud_run_v2_service.api_eu.name
+  role     = "roles/run.invoker"
+  member   = "allUsers"
+}
+
+resource "google_cloud_run_v2_service_iam_member" "api_apac_public" {
+  project  = var.project_id
+  location = "asia-northeast1"
+  name     = google_cloud_run_v2_service.api_apac.name
+  role     = "roles/run.invoker"
+  member   = "allUsers"
 }
diff --git a/servers/gcp/variables.tf b/servers/gcp/variables.tf
index dae6e0b..95119cf 100644
--- a/servers/gcp/variables.tf
+++ b/servers/gcp/variables.tf
@@ -5,7 +5,19 @@ variable "project_id" {
 }
 
 variable "image_tag" {
-  description = "Docker image tag to deploy"
+  description = "Docker image tag to deploy for the marketing site"
+  type        = string
+  default     = "latest"
+}
+
+variable "accounts_image_tag" {
+  description = "Docker image tag to deploy for the accounts service"
+  type        = string
+  default     = "latest"
+}
+
+variable "api_image_tag" {
+  description = "Docker image tag to deploy for the REST API (neuron-rest)"
   type        = string
   default     = "latest"
 }
@@ -20,23 +32,14 @@ variable "cloudflare_zone_id_neurontechnologies" {
 
 locals {
   project_id = var.project_id
-  image      = "us-central1-docker.pkg.dev/${var.project_id}/neuron-marketing/marketing:${var.image_tag}"
 
-  # Production secrets from GCP Secret Manager
-  secrets = [
-    "stripe-secret-key",
-    "stripe-webhook-secret",
-    "stripe-price-professional",
-    "stripe-price-founding",
-  ]
+  # Image refs per service
+  marketing_image = "us-central1-docker.pkg.dev/${var.project_id}/neuron-marketing/marketing:${var.image_tag}"
+  accounts_image  = "us-central1-docker.pkg.dev/${var.project_id}/neuron-accounts/accounts:${var.accounts_image_tag}"
+  api_image       = "us-central1-docker.pkg.dev/${var.project_id}/neuron-api/api:${var.api_image_tag}"
 
-  # Env var names for each secret (uppercase, hyphen → underscore)
-  secret_env_map = {
-    "stripe-secret-key"        = "STRIPE_SECRET_KEY"
-    "stripe-webhook-secret"    = "STRIPE_WEBHOOK_SECRET"
-    "stripe-price-professional" = "STRIPE_PRICE_PROFESSIONAL"
-    "stripe-price-founding"    = "STRIPE_PRICE_FOUNDING"
-  }
+  # Keep backward-compat alias (used in cloud-run.tf for marketing service)
+  image = local.marketing_image
 
   # Static publishable key (not sensitive — lives in source)
   stripe_publishable_key = "pk_live_51TPoHnJg9Fv1D3AUPMXnYyOJIVhn1FyH56zCMNnATo9tR7pO9lNDbXncp6VeDxm38qSdBHZPfqEBipUh3GZsWNyd00jvIO97E6"
diff --git a/servers/legion/apps/alloy.yaml b/servers/legion/apps/alloy.yaml
index 77dbfc0..245b88d 100644
--- a/servers/legion/apps/alloy.yaml
+++ b/servers/legion/apps/alloy.yaml
@@ -8,7 +8,7 @@ spec:
   source:
     repoURL: https://grafana.github.io/helm-charts
     chart: alloy
-    targetRevision: "*"
+    targetRevision: ">=0.9.0, <1.0.0"
     helm:
       values: |
         controller:
diff --git a/servers/legion/apps/gitea-runner.yaml b/servers/legion/apps/gitea-runner.yaml
index 2902f60..29ff958 100644
--- a/servers/legion/apps/gitea-runner.yaml
+++ b/servers/legion/apps/gitea-runner.yaml
@@ -1,3 +1,7 @@
+---
+# Gitea CI runner — general-purpose (legion)
+# Docker socket REMOVED. Builds go through buildkitd TCP endpoint.
+# Set DOCKER_HOST=tcp://buildkitd.ci.svc.cluster.local:1234 in build steps.
 apiVersion: apps/v1
 kind: Deployment
 metadata:
@@ -15,8 +19,10 @@ spec:
       labels:
         app: gitea-runner
       annotations:
-        config-version: "2026-04-23-nodeport-url"
+        config-version: "2026-04-25-buildkit-no-sock"
     spec:
+      securityContext:
+        runAsNonRoot: false  # act_runner needs root for container management
       initContainers:
         - name: register
           image: registry.neuralplatform.ai/ci-base:latest
@@ -36,10 +42,9 @@ spec:
                 timeout: 3h
               container:
                 network: host
-                docker_host: "unix:///var/run/docker.sock"
+                docker_host: "tcp://buildkitd.ci.svc.cluster.local:1234"
                 force_pull: false
-                valid_volumes:
-                  - /var/run/docker.sock
+                valid_volumes: []
                 default_image: "registry.neuralplatform.ai/ci-base:latest"
                 extra_hosts:
                   - "gitea.git.svc.cluster.local:10.43.1.53"
@@ -55,14 +60,16 @@ spec:
           image: registry.neuralplatform.ai/ci-base:latest
           workingDir: /data
           command: ["act_runner", "daemon", "--config", "/data/config.yaml"]
+          env:
+            - name: DOCKER_HOST
+              value: "tcp://buildkitd.ci.svc.cluster.local:1234"
           envFrom:
             - secretRef:
                 name: gitea-runner-secret
           volumeMounts:
             - name: data
               mountPath: /data
-            - name: docker-sock
-              mountPath: /var/run/docker.sock
+          # docker-sock volume intentionally removed — use buildkitd TCP instead
           resources:
             requests:
               memory: 512Mi
@@ -73,11 +80,9 @@ spec:
       volumes:
         - name: data
           emptyDir: {}
-        - name: docker-sock
-          hostPath:
-            path: /var/run/docker.sock
-            type: Socket
+        # docker-sock hostPath intentionally removed
 ---
+# Neuron Technologies CI runner
 apiVersion: apps/v1
 kind: Deployment
 metadata:
@@ -95,7 +100,7 @@ spec:
       labels:
         app: neuron-technologies-runner
       annotations:
-        config-version: "2026-04-24-initial"
+        config-version: "2026-04-25-buildkit-no-sock"
     spec:
       initContainers:
         - name: register
@@ -116,10 +121,9 @@ spec:
                 timeout: 3h
               container:
                 network: host
-                docker_host: "unix:///var/run/docker.sock"
+                docker_host: "tcp://buildkitd.ci.svc.cluster.local:1234"
                 force_pull: false
-                valid_volumes:
-                  - /var/run/docker.sock
+                valid_volumes: []
                 default_image: "registry.neuralplatform.ai/ci-base:latest"
                 extra_hosts:
                   - "gitea.git.svc.cluster.local:10.43.1.53"
@@ -135,14 +139,15 @@ spec:
           image: registry.neuralplatform.ai/ci-base:latest
           workingDir: /data
           command: ["act_runner", "daemon", "--config", "/data/config.yaml"]
+          env:
+            - name: DOCKER_HOST
+              value: "tcp://buildkitd.ci.svc.cluster.local:1234"
           envFrom:
             - secretRef:
                 name: neuron-technologies-runner-secret
           volumeMounts:
             - name: data
               mountPath: /data
-            - name: docker-sock
-              mountPath: /var/run/docker.sock
           resources:
             requests:
               memory: 512Mi
@@ -153,7 +158,3 @@ spec:
       volumes:
         - name: data
           emptyDir: {}
-        - name: docker-sock
-          hostPath:
-            path: /var/run/docker.sock
-            type: Socket
diff --git a/servers/legion/apps/kube-prometheus-stack.yaml b/servers/legion/apps/kube-prometheus-stack.yaml
index b3ff096..a584459 100644
--- a/servers/legion/apps/kube-prometheus-stack.yaml
+++ b/servers/legion/apps/kube-prometheus-stack.yaml
@@ -8,7 +8,7 @@ spec:
   source:
     repoURL: https://prometheus-community.github.io/helm-charts
     chart: kube-prometheus-stack
-    targetRevision: "*"
+    targetRevision: ">=60.0.0, <70.0.0"
     helm:
       values: |
         grafana:
@@ -37,6 +37,13 @@ spec:
               cert-manager.io/cluster-issuer: letsencrypt-prod
         prometheus:
           prometheusSpec:
+            # Discover ServiceMonitors and PrometheusRules from ALL namespaces.
+            # Without this, Prometheus only sees resources in the monitoring namespace.
+            serviceMonitorSelectorNilUsesHelmValues: false
+            serviceMonitorNamespaceSelector: {}
+            serviceMonitorSelector: {}
+            ruleNamespaceSelector: {}
+            podMonitorNamespaceSelector: {}
             storageSpec:
               volumeClaimTemplate:
                 spec:
diff --git a/servers/legion/apps/loki.yaml b/servers/legion/apps/loki.yaml
index 9105f7b..e9891ad 100644
--- a/servers/legion/apps/loki.yaml
+++ b/servers/legion/apps/loki.yaml
@@ -8,7 +8,7 @@ spec:
   source:
     repoURL: https://grafana.github.io/helm-charts
     chart: loki
-    targetRevision: "*"
+    targetRevision: ">=6.0.0, <7.0.0"
     helm:
       values: |
         loki:
diff --git a/servers/legion/apps/neuron-prod-project.yaml b/servers/legion/apps/neuron-prod-project.yaml
new file mode 100644
index 0000000..ec3e501
--- /dev/null
+++ b/servers/legion/apps/neuron-prod-project.yaml
@@ -0,0 +1,89 @@
+---
+# ArgoCD AppProject — neuron-prod
+# Restricts which repos can deploy to which namespaces.
+# Prevents misconfigured apps from deploying cluster-admin RBAC or
+# accidentally targeting personal/hobby namespaces.
+apiVersion: argoproj.io/v1alpha1
+kind: AppProject
+metadata:
+  name: neuron-prod
+  namespace: argocd
+  finalizers:
+    - resources-finalizer.argocd.argoproj.io
+spec:
+  description: "Neuron Technologies production services"
+
+  # Only allow deployments sourced from the infrastructure repo.
+  sourceRepos:
+    - "http://10.43.1.53:3000/will/infrastructure.git"
+    - "https://code.forgejo.org"
+    - "registry.neuralplatform.ai"
+
+  # Only allow deploying to neuron-prod and platform namespaces.
+  destinations:
+    - server: https://kubernetes.default.svc
+      namespace: neuron-prod
+    - server: https://kubernetes.default.svc
+      namespace: platform
+
+  # Cluster-scoped resources this project may NOT create.
+  clusterResourceBlacklist:
+    - group: rbac.authorization.k8s.io
+      kind: ClusterRole
+    - group: rbac.authorization.k8s.io
+      kind: ClusterRoleBinding
+    - group: ""
+      kind: Namespace
+
+  # Namespace-scoped resources this project may create.
+  namespaceResourceWhitelist:
+    - group: "apps"
+      kind: Deployment
+    - group: "apps"
+      kind: ReplicaSet
+    - group: "batch"
+      kind: CronJob
+    - group: "batch"
+      kind: Job
+    - group: ""
+      kind: Service
+    - group: ""
+      kind: ConfigMap
+    - group: ""
+      kind: Secret
+    - group: ""
+      kind: PersistentVolumeClaim
+    - group: ""
+      kind: ServiceAccount
+    - group: "networking.k8s.io"
+      kind: Ingress
+    - group: "networking.k8s.io"
+      kind: NetworkPolicy
+    - group: "policy"
+      kind: PodDisruptionBudget
+    - group: "autoscaling"
+      kind: HorizontalPodAutoscaler
+    - group: "traefik.io"
+      kind: IngressRoute
+    - group: "external-secrets.io"
+      kind: ExternalSecret
+    - group: "monitoring.coreos.com"
+      kind: ServiceMonitor
+    - group: "monitoring.coreos.com"
+      kind: PrometheusRule
+
+  # Sync windows: only allow automated syncs during low-traffic hours.
+  # Manual syncs allowed any time.
+  syncWindows:
+    - kind: allow
+      schedule: "0 2-6 * * *"   # 2am-6am UTC (off-peak)
+      duration: 4h
+      applications:
+        - "*"
+      manualSync: true
+    - kind: allow
+      schedule: "* * * * *"
+      duration: "1m"
+      manualSync: true           # Always allow manual deploys
+      applications:
+        - "*"
diff --git a/servers/legion/apps/postgres.yaml b/servers/legion/apps/postgres.yaml
index 2480f96..04ed5fd 100644
--- a/servers/legion/apps/postgres.yaml
+++ b/servers/legion/apps/postgres.yaml
@@ -9,7 +9,7 @@ spec:
     # Helm chart from Docker Hub OCI registry
     - repoURL: registry-1.docker.io/bitnamicharts
       chart: postgresql
-      targetRevision: "*"
+      targetRevision: ">=16.0.0, <17.0.0"
       helm:
         values: |
           auth:
diff --git a/servers/legion/apps/redis.yaml b/servers/legion/apps/redis.yaml
index d56c5e9..6d91a50 100644
--- a/servers/legion/apps/redis.yaml
+++ b/servers/legion/apps/redis.yaml
@@ -8,7 +8,7 @@ spec:
   source:
     repoURL: registry-1.docker.io/bitnamicharts
     chart: redis
-    targetRevision: "*"
+    targetRevision: ">=20.0.0, <21.0.0"
     helm:
       values: |
         auth:
diff --git a/servers/legion/apps/tempo.yaml b/servers/legion/apps/tempo.yaml
index 6c86318..dc35229 100644
--- a/servers/legion/apps/tempo.yaml
+++ b/servers/legion/apps/tempo.yaml
@@ -8,7 +8,7 @@ spec:
   source:
     repoURL: https://grafana.github.io/helm-charts
     chart: tempo
-    targetRevision: "*"
+    targetRevision: ">=1.0.0, <2.0.0"
     helm:
       values: |
         tempo:
diff --git a/servers/legion/k8s/backup/cronjob.yaml b/servers/legion/k8s/backup/cronjob.yaml
index 283d943..9a5367e 100644
--- a/servers/legion/k8s/backup/cronjob.yaml
+++ b/servers/legion/k8s/backup/cronjob.yaml
@@ -43,8 +43,10 @@ spec:
                 limits:
                   memory: 256Mi
           containers:
+            # Pin to explicit version — never use latest for backup tooling.
+            # Bump this intentionally after testing; do not let it float.
             - name: backup
-              image: restic/restic:latest
+              image: restic/restic:0.17.3
               command:
                 - /bin/sh
                 - -c
@@ -93,3 +95,77 @@ spec:
                 readOnly: true
             - name: dump
               emptyDir: {}
+---
+# Weekly restore-verify job — proves the backup is actually restorable.
+# Runs Sunday 4am. Restores latest snapshot to /tmp, validates the SQL dump.
+# Alerts via Alertmanager if it fails (BackupJobFailed PrometheusRule watches git namespace).
+apiVersion: batch/v1
+kind: CronJob
+metadata:
+  name: backup-verify
+  namespace: git
+spec:
+  schedule: "0 4 * * 0"
+  concurrencyPolicy: Forbid
+  failedJobsHistoryLimit: 2
+  successfulJobsHistoryLimit: 1
+  jobTemplate:
+    spec:
+      backoffLimit: 0
+      activeDeadlineSeconds: 1800  # 30 min max
+      template:
+        spec:
+          restartPolicy: Never
+          containers:
+            - name: verify
+              image: restic/restic:0.17.3
+              command:
+                - /bin/sh
+                - -c
+                - |
+                  set -e
+
+                  echo "=== Backup Restore Verification ==="
+                  echo "Listing latest snapshots..."
+                  restic snapshots --latest 3
+
+                  echo "Restoring latest snapshot to /tmp/restore-verify..."
+                  mkdir -p /tmp/restore-verify
+                  restic restore latest \
+                    --target /tmp/restore-verify \
+                    --include /all-databases.sql
+
+                  SQL_FILE="/tmp/restore-verify/all-databases.sql"
+                  if [ ! -f "$SQL_FILE" ]; then
+                    echo "ERROR: SQL dump not found in restored snapshot"
+                    exit 1
+                  fi
+
+                  SIZE=$(du -sh "$SQL_FILE" | cut -f1)
+                  echo "SQL dump restored: $SIZE"
+
+                  # Basic structural validation
+                  if ! grep -q "PostgreSQL database cluster dump" "$SQL_FILE"; then
+                    echo "ERROR: SQL dump does not look like a valid pg_dumpall output"
+                    exit 1
+                  fi
+
+                  TABLE_COUNT=$(grep -c "^CREATE TABLE" "$SQL_FILE" || true)
+                  echo "Tables found in dump: $TABLE_COUNT"
+
+                  if [ "$TABLE_COUNT" -lt 5 ]; then
+                    echo "ERROR: Too few tables in dump (expected >= 5, got $TABLE_COUNT)"
+                    exit 1
+                  fi
+
+                  echo "=== Verification PASSED: backup is restorable ==="
+                  rm -rf /tmp/restore-verify
+              envFrom:
+                - secretRef:
+                    name: backup-credentials
+              resources:
+                requests:
+                  memory: 256Mi
+                  cpu: 100m
+                limits:
+                  memory: 512Mi
diff --git a/servers/legion/k8s/external-secrets/cluster-secret-store.yaml b/servers/legion/k8s/external-secrets/cluster-secret-store.yaml
index 4c16fdf..5047b62 100644
--- a/servers/legion/k8s/external-secrets/cluster-secret-store.yaml
+++ b/servers/legion/k8s/external-secrets/cluster-secret-store.yaml
@@ -1,11 +1,47 @@
 ---
-# ClusterSecretStore — connects ESO to Vault using the root token Secret.
-# The vault-token Secret (name: vault-token, key: token) must be created
-# via kubectl as a one-time bootstrap step before this store can sync:
+# ClusterSecretStore — connects ESO to Vault
 #
-#   kubectl create secret generic vault-token \
-#     --namespace external-secrets \
-#     --from-literal=token=$(cat ~/Secrets/tokens/vault-root-token)
+# CURRENT STATE: token auth (root token). This works but the root token
+# has no expiry and full Vault access — it's a single credential that
+# compromises everything if leaked.
+#
+# MIGRATION TO APPROLE (do this before GCP migration):
+# ─────────────────────────────────────────────────────
+# 1. Enable AppRole on Vault (if not already):
+#      vault auth enable approle
+#
+# 2. Create a policy scoped to only the paths ESO reads:
+#      vault policy write eso-read - <<EOF
+#      path "secret/data/neuron-technologies/*" { capabilities = ["read"] }
+#      path "secret/data/r2"                    { capabilities = ["read"] }
+#      path "secret/data/slack"                 { capabilities = ["read"] }
+#      path "secret/data/legion-db"             { capabilities = ["read"] }
+#      EOF
+#
+# 3. Create the AppRole:
+#      vault write auth/approle/role/eso \
+#        token_policies="eso-read" \
+#        token_ttl=1h \
+#        token_max_ttl=4h \
+#        secret_id_ttl=0  # no expiry for the secretId
+#
+# 4. Get the roleId and secretId:
+#      vault read auth/approle/role/eso/role-id    → ROLE_ID
+#      vault write -f auth/approle/role/eso/secret-id → SECRET_ID
+#
+# 5. Create the k8s secret (one-time bootstrap, same as the root token was):
+#      kubectl create secret generic vault-approle \
+#        --namespace external-secrets \
+#        --from-literal=roleId=<ROLE_ID> \
+#        --from-literal=secretId=<SECRET_ID>
+#
+# 6. Flip this file to the appRole stanza below, commit, push.
+#    ESO will hot-reload when the manifest changes.
+#
+# 7. Delete the old vault-token secret:
+#      kubectl delete secret vault-token -n external-secrets
+# ─────────────────────────────────────────────────────
+
 apiVersion: external-secrets.io/v1beta1
 kind: ClusterSecretStore
 metadata:
@@ -17,7 +53,17 @@ spec:
       path: "secret"
       version: "v2"
       auth:
+        # ── Current: root token auth ──────────────────────────────────────────
+        # TODO: replace with appRole stanza once AppRole is bootstrapped (see above).
         tokenSecretRef:
           name: vault-token
           namespace: external-secrets
           key: token
+        # ── Target: AppRole auth (uncomment after bootstrap) ──────────────────
+        # appRole:
+        #   path: "approle"
+        #   roleId: "eso"
+        #   secretRef:
+        #     name: vault-approle
+        #     namespace: external-secrets
+        #     key: secretId
diff --git a/servers/legion/k8s/gitea-runner/buildkitd.yaml b/servers/legion/k8s/gitea-runner/buildkitd.yaml
new file mode 100644
index 0000000..7abb889
--- /dev/null
+++ b/servers/legion/k8s/gitea-runner/buildkitd.yaml
@@ -0,0 +1,71 @@
+---
+# BuildKit daemon — rootless image builder for CI runners.
+# Replaces the /var/run/docker.sock hostPath mount that gave containers
+# full root access to the Legion host node.
+#
+# Runners connect via TCP: DOCKER_HOST=tcp://buildkitd.ci.svc.cluster.local:1234
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: buildkitd
+  namespace: ci
+  labels:
+    app: buildkitd
+spec:
+  replicas: 1
+  selector:
+    matchLabels:
+      app: buildkitd
+  template:
+    metadata:
+      labels:
+        app: buildkitd
+    spec:
+      containers:
+        - name: buildkitd
+          image: moby/buildkit:v0.19.0-rootless
+          args:
+            - --oci-worker-no-process-sandbox
+            - --addr
+            - tcp://0.0.0.0:1234
+            - --addr
+            - unix:///run/buildkit/buildkitd.sock
+          ports:
+            - name: tcp
+              containerPort: 1234
+          securityContext:
+            seccompProfile:
+              type: Unconfined       # BuildKit rootless requires this
+            runAsUser: 1000
+            runAsGroup: 1000
+          readinessProbe:
+            exec:
+              command: ["buildctl", "debug", "workers"]
+            initialDelaySeconds: 5
+            periodSeconds: 10
+          resources:
+            requests:
+              cpu: 200m
+              memory: 256Mi
+            limits:
+              cpu: "4"
+              memory: 4Gi
+          volumeMounts:
+            - name: buildkit-storage
+              mountPath: /home/user/.local/share/buildkit
+      volumes:
+        - name: buildkit-storage
+          emptyDir: {}
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: buildkitd
+  namespace: ci
+spec:
+  selector:
+    app: buildkitd
+  ports:
+    - name: tcp
+      port: 1234
+      targetPort: 1234
diff --git a/servers/legion/k8s/monitoring/alerting-rules.yaml b/servers/legion/k8s/monitoring/alerting-rules.yaml
new file mode 100644
index 0000000..96382fd
--- /dev/null
+++ b/servers/legion/k8s/monitoring/alerting-rules.yaml
@@ -0,0 +1,153 @@
+---
+# PrometheusRules — SLO and operational alerts for neuron-prod
+# These fire into Alertmanager which routes to #infrastructure-alerts on Slack.
+apiVersion: monitoring.coreos.com/v1
+kind: PrometheusRule
+metadata:
+  name: neuron-slo
+  namespace: monitoring
+  labels:
+    release: kube-prometheus-stack
+spec:
+  groups:
+
+    # ── Neuron API SLOs ──────────────────────────────────────��───────────────
+    - name: neuron.slo
+      interval: 60s
+      rules:
+        - alert: NeuronMCPHighErrorRate
+          expr: |
+            (
+              sum(rate(http_server_requests_seconds_count{namespace="neuron-prod",app="neuron-mcp",status=~"5.."}[5m]))
+              /
+              sum(rate(http_server_requests_seconds_count{namespace="neuron-prod",app="neuron-mcp"}[5m]))
+            ) > 0.01
+          for: 2m
+          labels:
+            severity: critical
+            namespace: neuron-prod
+          annotations:
+            summary: "neuron-mcp 5xx error rate > 1%"
+            description: "Error rate is {{ $value | humanizePercentage }} over the last 5 minutes."
+
+        - alert: NeuronRestHighErrorRate
+          expr: |
+            (
+              sum(rate(http_server_requests_seconds_count{namespace="neuron-prod",app="neuron-rest",status=~"5.."}[5m]))
+              /
+              sum(rate(http_server_requests_seconds_count{namespace="neuron-prod",app="neuron-rest"}[5m]))
+            ) > 0.01
+          for: 2m
+          labels:
+            severity: critical
+            namespace: neuron-prod
+          annotations:
+            summary: "neuron-rest 5xx error rate > 1%"
+            description: "Error rate is {{ $value | humanizePercentage }} over the last 5 minutes."
+
+        - alert: NeuronMCPHighLatency
+          expr: |
+            histogram_quantile(0.99,
+              sum(rate(http_server_requests_seconds_bucket{namespace="neuron-prod",app="neuron-mcp"}[10m]))
+              by (le)
+            ) > 2.0
+          for: 5m
+          labels:
+            severity: warning
+            namespace: neuron-prod
+          annotations:
+            summary: "neuron-mcp p99 latency > 2s"
+            description: "p99 latency is {{ $value | humanizeDuration }} over the last 10 minutes."
+
+        - alert: NeuronRestHighLatency
+          expr: |
+            histogram_quantile(0.99,
+              sum(rate(http_server_requests_seconds_bucket{namespace="neuron-prod",app="neuron-rest"}[10m]))
+              by (le)
+            ) > 2.0
+          for: 5m
+          labels:
+            severity: warning
+            namespace: neuron-prod
+          annotations:
+            summary: "neuron-rest p99 latency > 2s"
+            description: "p99 latency is {{ $value | humanizeDuration }} over the last 10 minutes."
+
+    # ── Pod health ───────────────────────────────────────────────────────────
+    - name: neuron.pods
+      rules:
+        - alert: NeuronPodCrashLooping
+          expr: |
+            rate(kube_pod_container_status_restarts_total{namespace="neuron-prod"}[15m]) * 60 * 15 > 3
+          for: 5m
+          labels:
+            severity: critical
+            namespace: neuron-prod
+          annotations:
+            summary: "Pod {{ $labels.pod }} is crash looping"
+            description: "Container {{ $labels.container }} restarted {{ $value | humanize }} times in 15 minutes."
+
+        - alert: NeuronPodNotReady
+          expr: |
+            sum by (pod) (kube_pod_status_ready{namespace="neuron-prod",condition="false"}) > 0
+          for: 5m
+          labels:
+            severity: warning
+            namespace: neuron-prod
+          annotations:
+            summary: "Pod {{ $labels.pod }} not ready for 5 minutes"
+
+        - alert: NeuronDeploymentReplicasMissing
+          expr: |
+            kube_deployment_status_replicas_available{namespace="neuron-prod"}
+            < kube_deployment_spec_replicas{namespace="neuron-prod"}
+          for: 3m
+          labels:
+            severity: warning
+            namespace: neuron-prod
+          annotations:
+            summary: "Deployment {{ $labels.deployment }} has fewer replicas than desired"
+
+    # ── Postgres ─────────────────────────���─────────────────────────────────��─
+    - name: postgres.health
+      rules:
+        - alert: PostgresDown
+          expr: pg_up{namespace="platform"} == 0
+          for: 2m
+          labels:
+            severity: critical
+          annotations:
+            summary: "Postgres is down"
+
+        - alert: PostgresHighConnections
+          expr: |
+            (pg_stat_database_numbackends{namespace="platform"} / pg_settings_max_connections{namespace="platform"})
+            > 0.80
+          for: 5m
+          labels:
+            severity: warning
+          annotations:
+            summary: "Postgres connection pool > 80%"
+            description: "Connection utilization is {{ $value | humanizePercentage }}."
+
+    # ── Backup health ────────────────────────────���───────────────────────────
+    - name: backup.health
+      rules:
+        - alert: BackupJobFailed
+          expr: |
+            kube_job_status_failed{namespace="git",job_name=~"gitea-backup.*"} > 0
+          for: 0m
+          labels:
+            severity: critical
+          annotations:
+            summary: "Gitea backup job failed"
+            description: "Check: kubectl logs -n git -l job-name=gitea-backup --previous"
+
+        - alert: BackupNotRunRecently
+          expr: |
+            (time() - kube_job_status_completion_time{namespace="git",job_name=~"gitea-backup.*"}) > 90000
+          for: 0m
+          labels:
+            severity: warning
+          annotations:
+            summary: "Last successful backup was more than 25 hours ago"
diff --git a/servers/legion/k8s/monitoring/servicemonitor.yaml b/servers/legion/k8s/monitoring/servicemonitor.yaml
new file mode 100644
index 0000000..2ac9c2d
--- /dev/null
+++ b/servers/legion/k8s/monitoring/servicemonitor.yaml
@@ -0,0 +1,64 @@
+---
+# ServiceMonitors for neuron-prod — scraped by kube-prometheus-stack Prometheus
+# Requires: kube-prometheus-stack prometheusSpec.serviceMonitorSelectorNilUsesHelmValues=false
+# OR a serviceMonitorNamespaceSelector that includes neuron-prod.
+# We set this in the Prometheus stack values (see apps/kube-prometheus-stack.yaml).
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: neuron-mcp
+  namespace: monitoring
+  labels:
+    release: kube-prometheus-stack
+spec:
+  namespaceSelector:
+    matchNames:
+      - neuron-prod
+  selector:
+    matchLabels:
+      app: neuron-mcp
+  endpoints:
+    - port: http
+      path: /actuator/prometheus
+      interval: 30s
+      scrapeTimeout: 10s
+---
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: neuron-rest
+  namespace: monitoring
+  labels:
+    release: kube-prometheus-stack
+spec:
+  namespaceSelector:
+    matchNames:
+      - neuron-prod
+  selector:
+    matchLabels:
+      app: neuron-rest
+  endpoints:
+    - port: http
+      path: /actuator/prometheus
+      interval: 30s
+      scrapeTimeout: 10s
+---
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: neuron-marketing
+  namespace: monitoring
+  labels:
+    release: kube-prometheus-stack
+spec:
+  namespaceSelector:
+    matchNames:
+      - neuron-prod
+  selector:
+    matchLabels:
+      app: neuron-marketing
+  endpoints:
+    - port: http
+      path: /api/metrics
+      interval: 60s
+      scrapeTimeout: 15s
diff --git a/servers/legion/k8s/neuron-technologies/marketing/deployment.yaml b/servers/legion/k8s/neuron-technologies/marketing/deployment.yaml
index 1c7d775..53a973c 100644
--- a/servers/legion/k8s/neuron-technologies/marketing/deployment.yaml
+++ b/servers/legion/k8s/neuron-technologies/marketing/deployment.yaml
@@ -17,6 +17,11 @@ spec:
         app: neuron-marketing
         env: prod
     spec:
+      securityContext:
+        runAsNonRoot: true
+        runAsUser: 1001
+        seccompProfile:
+          type: RuntimeDefault
       containers:
         - name: neuron-marketing
           image: registry.neuralplatform.ai/neuron-technologies/marketing:1e94e8ae
@@ -24,6 +29,12 @@ spec:
           ports:
             - name: http
               containerPort: 3000
+          securityContext:
+            allowPrivilegeEscalation: false
+            readOnlyRootFilesystem: false  # Next.js writes .next/cache at runtime
+            runAsNonRoot: true
+            capabilities:
+              drop: ["ALL"]
           env:
             - name: NODE_ENV
               value: production
diff --git a/servers/legion/k8s/neuron-technologies/prod/hpa.yaml b/servers/legion/k8s/neuron-technologies/prod/hpa.yaml
new file mode 100644
index 0000000..b947c72
--- /dev/null
+++ b/servers/legion/k8s/neuron-technologies/prod/hpa.yaml
@@ -0,0 +1,98 @@
+---
+# HorizontalPodAutoscalers for neuron-prod
+# Single Legion node today — minReplicas=1 prevents unnecessary idle waste.
+# When GCP GKE is live these minimums should go to 2 for HA.
+apiVersion: autoscaling/v2
+kind: HorizontalPodAutoscaler
+metadata:
+  name: neuron-mcp-hpa
+  namespace: neuron-prod
+spec:
+  scaleTargetRef:
+    apiVersion: apps/v1
+    kind: Deployment
+    name: neuron-mcp-blue
+  minReplicas: 1
+  maxReplicas: 6
+  metrics:
+    - type: Resource
+      resource:
+        name: cpu
+        target:
+          type: Utilization
+          averageUtilization: 65
+    - type: Resource
+      resource:
+        name: memory
+        target:
+          type: Utilization
+          averageUtilization: 80
+  behavior:
+    scaleDown:
+      stabilizationWindowSeconds: 300  # Don't thrash — wait 5m before scaling down
+      policies:
+        - type: Pods
+          value: 1
+          periodSeconds: 120
+    scaleUp:
+      stabilizationWindowSeconds: 30
+      policies:
+        - type: Pods
+          value: 2
+          periodSeconds: 60
+---
+apiVersion: autoscaling/v2
+kind: HorizontalPodAutoscaler
+metadata:
+  name: neuron-rest-hpa
+  namespace: neuron-prod
+spec:
+  scaleTargetRef:
+    apiVersion: apps/v1
+    kind: Deployment
+    name: neuron-rest
+  minReplicas: 1
+  maxReplicas: 4
+  metrics:
+    - type: Resource
+      resource:
+        name: cpu
+        target:
+          type: Utilization
+          averageUtilization: 65
+    - type: Resource
+      resource:
+        name: memory
+        target:
+          type: Utilization
+          averageUtilization: 80
+  behavior:
+    scaleDown:
+      stabilizationWindowSeconds: 300
+    scaleUp:
+      stabilizationWindowSeconds: 30
+---
+apiVersion: autoscaling/v2
+kind: HorizontalPodAutoscaler
+metadata:
+  name: neuron-marketing-hpa
+  namespace: neuron-prod
+spec:
+  scaleTargetRef:
+    apiVersion: apps/v1
+    kind: Deployment
+    name: neuron-marketing
+  minReplicas: 2
+  maxReplicas: 8
+  metrics:
+    - type: Resource
+      resource:
+        name: cpu
+        target:
+          type: Utilization
+          averageUtilization: 70
+  behavior:
+    scaleDown:
+      stabilizationWindowSeconds: 120
+    scaleUp:
+      stabilizationWindowSeconds: 15
diff --git a/servers/legion/k8s/neuron-technologies/prod/kustomization.yaml b/servers/legion/k8s/neuron-technologies/prod/kustomization.yaml
index 3d1ae8b..5c73cd9 100644
--- a/servers/legion/k8s/neuron-technologies/prod/kustomization.yaml
+++ b/servers/legion/k8s/neuron-technologies/prod/kustomization.yaml
@@ -15,3 +15,7 @@ resources:
   - backup-cronjob.yaml
   - runpod-inference.yaml
   - runpod-lb-configmap.yaml
+  # Hardening additions
+  - hpa.yaml
+  - pdb.yaml
+  - network-policy.yaml
diff --git a/servers/legion/k8s/neuron-technologies/prod/mcp-deployment.yaml b/servers/legion/k8s/neuron-technologies/prod/mcp-deployment.yaml
index 9215924..05769ef 100644
--- a/servers/legion/k8s/neuron-technologies/prod/mcp-deployment.yaml
+++ b/servers/legion/k8s/neuron-technologies/prod/mcp-deployment.yaml
@@ -25,6 +25,10 @@ spec:
         kubectl.kubernetes.io/restartedAt: "2026-04-25T08:40:00Z"
     spec:
       terminationGracePeriodSeconds: 30
+      securityContext:
+        runAsNonRoot: true
+        seccompProfile:
+          type: RuntimeDefault
       containers:
         - name: neuron-mcp
           image: registry.neuralplatform.ai/neuron-technologies/neuron-mcp:v0.15.3
@@ -32,6 +36,12 @@ spec:
           ports:
             - name: http
               containerPort: 8080
+          securityContext:
+            allowPrivilegeEscalation: false
+            readOnlyRootFilesystem: false  # JVM writes temp files; restrict paths via allowedPaths once profiled
+            runAsNonRoot: true
+            capabilities:
+              drop: ["ALL"]
           env:
             - name: OTEL_EXPORTER_OTLP_ENDPOINT
               value: "http://alloy-otlp.monitoring.svc.cluster.local:4318"
@@ -57,18 +67,25 @@ spec:
               exec:
                 # Give Traefik time to drain in-flight connections before pod terminates
                 command: ["sh", "-c", "sleep 8"]
+          # startupProbe gives the JVM up to 150s to start before liveness takes over.
+          # Prevents false crash loops during slow cold starts.
+          startupProbe:
+            httpGet:
+              path: /actuator/health/liveness
+              port: http
+            failureThreshold: 30
+            periodSeconds: 5
           livenessProbe:
             httpGet:
               path: /actuator/health/liveness
               port: http
-            initialDelaySeconds: 30
             periodSeconds: 30
             failureThreshold: 3
           readinessProbe:
             httpGet:
               path: /actuator/health/readiness
               port: http
-            initialDelaySeconds: 10
+            initialDelaySeconds: 5
             periodSeconds: 10
             failureThreshold: 3
       volumes:
diff --git a/servers/legion/k8s/neuron-technologies/prod/network-policy.yaml b/servers/legion/k8s/neuron-technologies/prod/network-policy.yaml
new file mode 100644
index 0000000..7f9dec9
--- /dev/null
+++ b/servers/legion/k8s/neuron-technologies/prod/network-policy.yaml
@@ -0,0 +1,124 @@
+---
+# NetworkPolicies for neuron-prod
+# Default-deny all ingress and egress, then explicitly allow only what's needed.
+# Enforced by k3s's built-in NetworkPolicy controller.
+
+# ── Baseline: deny all traffic in namespace ───────────────────────────────────
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: default-deny-all
+  namespace: neuron-prod
+spec:
+  podSelector: {}
+  policyTypes:
+    - Ingress
+    - Egress
+---
+# ── neuron-mcp: accept traffic from Traefik (kube-system) only ───────────────
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: allow-mcp-ingress
+  namespace: neuron-prod
+spec:
+  podSelector:
+    matchLabels:
+      app: neuron-mcp
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: kube-system
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: neuron-prod
+---
+# ── neuron-rest: accept from mcp, kube-system (Traefik), monitoring (Alloy) ──
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: allow-rest-ingress
+  namespace: neuron-prod
+spec:
+  podSelector:
+    matchLabels:
+      app: neuron-rest
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: kube-system
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: neuron-prod
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: monitoring
+---
+# ── neuron-marketing: accept from Traefik only ───────────────────────────────
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: allow-marketing-ingress
+  namespace: neuron-prod
+spec:
+  podSelector:
+    matchLabels:
+      app: neuron-marketing
+  policyTypes:
+    - Ingress
+  ingress:
+    - from:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: kube-system
+---
+# ── Egress: all prod pods may reach platform (postgres/redis), vault,
+#            monitoring (alloy OTLP), kube-dns, and the internet (external APIs) ─
+apiVersion: networking.k8s.io/v1
+kind: NetworkPolicy
+metadata:
+  name: allow-prod-egress
+  namespace: neuron-prod
+spec:
+  podSelector: {}
+  policyTypes:
+    - Egress
+  egress:
+    # Kubernetes DNS
+    - ports:
+        - port: 53
+          protocol: UDP
+        - port: 53
+          protocol: TCP
+    # platform namespace (Postgres, Redis, accounts service)
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: platform
+    # vault namespace
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: vault
+    # monitoring namespace (Alloy OTLP)
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: monitoring
+    # intra-namespace (mcp ↔ rest ↔ license)
+    - to:
+        - namespaceSelector:
+            matchLabels:
+              kubernetes.io/metadata.name: neuron-prod
+    # external internet (Stripe, OAuth providers, Cloudflare, etc.)
+    - ports:
+        - port: 443
+          protocol: TCP
+        - port: 80
+          protocol: TCP
diff --git a/servers/legion/k8s/neuron-technologies/prod/pdb.yaml b/servers/legion/k8s/neuron-technologies/prod/pdb.yaml
new file mode 100644
index 0000000..a83438f
--- /dev/null
+++ b/servers/legion/k8s/neuron-technologies/prod/pdb.yaml
@@ -0,0 +1,39 @@
+---
+# PodDisruptionBudgets for neuron-prod
+# Ensures at least 1 replica of each service survives node drains and upgrades.
+# With replicas=1 today, minAvailable=1 means drain will block until HPA scales up
+# to 2+ — this is intentional and provides a natural safety gate.
+apiVersion: policy/v1
+kind: PodDisruptionBudget
+metadata:
+  name: neuron-mcp-pdb
+  namespace: neuron-prod
+spec:
+  minAvailable: 1
+  selector:
+    matchLabels:
+      app: neuron-mcp
+      slot: blue
+---
+apiVersion: policy/v1
+kind: PodDisruptionBudget
+metadata:
+  name: neuron-rest-pdb
+  namespace: neuron-prod
+spec:
+  minAvailable: 1
+  selector:
+    matchLabels:
+      app: neuron-rest
+---
+apiVersion: policy/v1
+kind: PodDisruptionBudget
+metadata:
+  name: neuron-marketing-pdb
+  namespace: neuron-prod
+spec:
+  # Marketing runs 2 replicas — keep at least 1 alive during drains.
+  minAvailable: 1
+  selector:
+    matchLabels:
+      app: neuron-marketing
diff --git a/servers/legion/k8s/neuron-technologies/prod/rest-deployment.yaml b/servers/legion/k8s/neuron-technologies/prod/rest-deployment.yaml
index 59c3508..4cdf765 100644
--- a/servers/legion/k8s/neuron-technologies/prod/rest-deployment.yaml
+++ b/servers/legion/k8s/neuron-technologies/prod/rest-deployment.yaml
@@ -19,6 +19,10 @@ spec:
       annotations:
         kubectl.kubernetes.io/restartedAt: "2026-04-24T10:00:00Z"
     spec:
+      securityContext:
+        runAsNonRoot: true
+        seccompProfile:
+          type: RuntimeDefault
       containers:
         - name: neuron-rest
           image: registry.neuralplatform.ai/neuron-technologies/neuron-rest:v0.15.3
@@ -26,6 +30,12 @@ spec:
           ports:
             - name: http
               containerPort: 8081
+          securityContext:
+            allowPrivilegeEscalation: false
+            readOnlyRootFilesystem: false
+            runAsNonRoot: true
+            capabilities:
+              drop: ["ALL"]
           envFrom:
             - configMapRef:
                 name: neuron-prod-config
@@ -41,18 +51,23 @@ spec:
             limits:
               cpu: 1000m
               memory: 1Gi
+          startupProbe:
+            httpGet:
+              path: /actuator/health/liveness
+              port: http
+            failureThreshold: 30
+            periodSeconds: 5
           livenessProbe:
             httpGet:
               path: /actuator/health/liveness
               port: http
-            initialDelaySeconds: 30
             periodSeconds: 30
             failureThreshold: 3
           readinessProbe:
             httpGet:
               path: /actuator/health/readiness
               port: http
-            initialDelaySeconds: 15
+            initialDelaySeconds: 5
             periodSeconds: 10
             failureThreshold: 3
       volumes:
diff --git a/servers/legion/scripts/neuron-self-improve.sh b/servers/legion/scripts/neuron-self-improve.sh
index 1a9d16d..5055544 100755
--- a/servers/legion/scripts/neuron-self-improve.sh
+++ b/servers/legion/scripts/neuron-self-improve.sh
@@ -1,46 +1,27 @@
 #!/usr/bin/env bash
 # neuron-self-improve.sh — self-improvement loop for Neuron
 #
-# Manages code experiments across alpha/beta/gamma zones and promotes winners to prod.
-# All cluster changes go through git → Argo CD. No direct kubectl mutations.
+# Model: stage is the experiment slot. prod runs blue/green.
+# Promote flow: implement → CI → stage → verify → blue-green-deploy → prod.
 #
 # Usage:
-#   ./neuron-self-improve.sh snapshot <slot>          # copy prod DB to slot
-#   ./neuron-self-improve.sh deploy <slot> <sha>      # point slot at a specific image
-#   ./neuron-self-improve.sh promote <slot>           # copy slot's image to prod
-#   ./neuron-self-improve.sh status                   # show what's running in each zone
-#   ./neuron-self-improve.sh rollback <slot> <sha>    # revert slot to a previous image
+#   ./neuron-self-improve.sh snapshot              # copy prod DB to stage
+#   ./neuron-self-improve.sh deploy <sha>          # point stage at a specific image
+#   ./neuron-self-improve.sh promote <tag>         # blue/green flip prod to <tag>
+#   ./neuron-self-improve.sh status                # show active blue/green slot + stage
+#   ./neuron-self-improve.sh rollback <tag>        # roll prod back to <tag> via blue/green
 
 set -euo pipefail
 
 INFRA_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-REGISTRY="registry.neuralplatform.ai/neural-platform/neuron"
+REGISTRY="registry.neuralplatform.ai/neuron-technologies/neuron-mcp"
 KUBECONFIG="${KUBECONFIG:-$HOME/.kube/legion-config}"
 
+KC_PROD="kubectl --kubeconfig=$KUBECONFIG -n neuron-prod"
+KC_STAGE="kubectl --kubeconfig=$KUBECONFIG -n neuron-stage"
+
 # ── helpers ───────────────────────────────────────────────────────────────────
 
-slot_manifest() {
-  local slot=$1
-  if [[ "$slot" == "prod" ]]; then
-    echo "$INFRA_DIR/servers/legion/apps/neuron.yaml"
-  else
-    echo "$INFRA_DIR/servers/legion/apps/neuron-${slot}.yaml"
-  fi
-}
-
-current_sha() {
-  local slot=$1
-  grep "image:" "$(slot_manifest "$slot")" | grep -oP '[a-f0-9]{40}' | head -1
-}
-
-set_image() {
-  local slot=$1 sha=$2
-  local manifest
-  manifest="$(slot_manifest "$slot")"
-  sed -i "s|${REGISTRY}:.*|${REGISTRY}:${sha}|" "$manifest"
-  echo "  set $slot → $sha"
-}
-
 git_push() {
   local msg=$1
   cd "$INFRA_DIR"
@@ -50,35 +31,42 @@ git_push() {
   echo "  pushed: $msg"
 }
 
-wait_healthy() {
-  local slot=$1
-  local app="neuron"
-  [[ "$slot" != "prod" ]] && app="neuron-${slot}"
-  echo "  waiting for $slot to be healthy..."
-  local i=0
-  while [[ $i -lt 30 ]]; do
-    local ready
-    ready=$(KUBECONFIG="$KUBECONFIG" kubectl get pods -n neuron -l "app=${app}" \
-      --no-headers 2>/dev/null | grep "1/1" | wc -l)
-    [[ "$ready" -ge 1 ]] && echo "  $slot is healthy" && return 0
-    sleep 10
-    ((i++))
-  done
-  echo "  WARNING: $slot did not become healthy within 5 minutes"
-  return 1
+active_prod_slot() {
+  $KC_PROD get svc neuron-mcp -o jsonpath='{.spec.selector.slot}' 2>/dev/null || echo "unknown"
+}
+
+current_prod_tag() {
+  local slot
+  slot=$(active_prod_slot)
+  $KC_PROD get deployment "neuron-mcp-$slot" \
+    -o jsonpath='{.spec.template.spec.containers[0].image}' 2>/dev/null \
+    | cut -d: -f2 || echo "unknown"
+}
+
+wait_ready() {
+  local ns=$1 label=$2 timeout=${3:-180}
+  echo "  waiting for $label to be ready..."
+  kubectl --kubeconfig="$KUBECONFIG" -n "$ns" \
+    rollout status deployment "$label" --timeout="${timeout}s"
+  echo "  $label is ready"
 }
 
 # ── commands ──────────────────────────────────────────────────────────────────
 
 cmd_snapshot() {
-  local slot=${1:?Usage: snapshot <slot>}
-  echo "==> Snapshotting prod DB into $slot..."
-  KUBECONFIG="$KUBECONFIG" kubectl apply -f - << EOF
+  echo "==> Snapshotting prod DB into stage..."
+  local prod_slot
+  prod_slot=$(active_prod_slot)
+  local prod_pvc="neuron-prod-data"
+  local stage_pvc="neuron-stage-data"
+
+  # Run a one-shot job that copies prod DB to stage PVC
+  $KC_PROD apply -f - << EOF
 apiVersion: batch/v1
 kind: Job
 metadata:
-  name: snapshot-prod-to-${slot}
-  namespace: neuron
+  name: snapshot-prod-to-stage
+  namespace: neuron-prod
 spec:
   ttlSecondsAfterFinished: 300
   template:
@@ -89,7 +77,9 @@ spec:
           image: keinos/sqlite3:latest
           command: ["/bin/sh", "-c"]
           args:
-            - sqlite3 /src/neuron.db ".backup /dst/neuron.db" && echo "Done" && ls -lh /dst/neuron.db
+            - |
+              sqlite3 /src/neuron.db ".backup /dst/neuron.db"
+              echo "Done — $(ls -lh /dst/neuron.db | awk '{print \$5}')"
           volumeMounts:
             - name: src
               mountPath: /src
@@ -99,125 +89,127 @@ spec:
       volumes:
         - name: src
           persistentVolumeClaim:
-            claimName: neuron-data
+            claimName: ${prod_pvc}
         - name: dst
           persistentVolumeClaim:
-            claimName: neuron-${slot}-data
+            claimName: ${stage_pvc}
 EOF
+
   echo "  waiting for snapshot job..."
-  KUBECONFIG="$KUBECONFIG" kubectl wait job "snapshot-prod-to-${slot}" \
-    -n neuron --for=condition=complete --timeout=120s
-  echo "  snapshot complete"
+  $KC_PROD wait job snapshot-prod-to-stage --for=condition=complete --timeout=120s
+  echo "  snapshot complete — stage DB is now a copy of prod"
 }
 
 cmd_deploy() {
-  local slot=${1:?Usage: deploy <slot> <sha>}
-  local sha=${2:?Usage: deploy <slot> <sha>}
-  echo "==> Deploying $sha to $slot..."
-  set_image "$slot" "$sha"
-  git_push "chore(neuron): deploy ${sha:0:12} to $slot"
-  wait_healthy "$slot"
-  echo "==> $slot is live at https://neuron-${slot}.neuralplatform.ai"
+  local sha=${1:?Usage: deploy <sha>}
+  echo "==> Deploying $sha to stage..."
+
+  # Update stage deployment image in-place (stage is a single deployment, not blue/green)
+  $KC_STAGE set image deployment/neuron-mcp "neuron-mcp=${REGISTRY}:${sha}"
+  $KC_STAGE rollout restart deployment/neuron-mcp
+  wait_ready "neuron-stage" "neuron-mcp"
+
+  echo "==> stage is live"
+  echo "  endpoint: https://neuron.neuralplatform.ai/stage  (or stage ingress if configured)"
+  echo "  MCP URL: (see stage ingress/services.yaml)"
 }
 
 cmd_promote() {
-  local slot=${1:?Usage: promote <slot>}
-  local sha
-  sha=$(current_sha "$slot")
-  [[ -z "$sha" ]] && echo "ERROR: could not read SHA from $slot" && exit 1
-  echo "==> Promoting $slot ($sha) to prod..."
-  set_image "prod" "$sha"
-  # Also stamp the :stable registry tag (pull and re-tag)
-  git_push "chore(neuron): promote $slot (${sha:0:12}) to prod"
-  wait_healthy "prod"
-  echo "==> prod is now running $sha"
-  echo "==> prod endpoint: https://neuron.neuralplatform.ai"
+  local tag=${1:?Usage: promote <tag>}
+  echo "==> Promoting $tag to prod via blue/green flip..."
+  "$INFRA_DIR/scripts/blue-green-deploy.sh" "$tag"
 }
 
 cmd_rollback() {
-  local slot=${1:?Usage: rollback <slot> <sha>}
-  local sha=${2:?Usage: rollback <slot> <sha>}
-  echo "==> Rolling $slot back to $sha..."
-  set_image "$slot" "$sha"
-  git_push "chore(neuron): rollback $slot to ${sha:0:12}"
-  wait_healthy "$slot"
-  echo "==> $slot rolled back"
+  local tag=${1:?Usage: rollback <tag>}
+  echo "==> Rolling prod back to $tag via blue/green..."
+  "$INFRA_DIR/scripts/blue-green-deploy.sh" "$tag"
 }
 
 cmd_status() {
-  echo "==> Neuron slot status"
+  echo "==> Neuron deployment status"
   echo ""
-  for slot in prod alpha beta gamma; do
-    local app="neuron"
-    [[ "$slot" != "prod" ]] && app="neuron-${slot}"
-    local sha
-    sha=$(current_sha "$slot" 2>/dev/null || echo "unknown")
-    local ready
-    ready=$(KUBECONFIG="$KUBECONFIG" kubectl get pods -n neuron -l "app=${app}" \
-      --no-headers 2>/dev/null | grep "1/1" | wc -l | tr -d ' ')
-    printf "  %-8s  %s  pods_ready=%s\n" "$slot" "${sha:0:12}" "$ready"
-  done
+
+  local active_slot
+  active_slot=$(active_prod_slot)
+  local prod_tag
+  prod_tag=$(current_prod_tag)
+
+  printf "  %-8s  slot=%-5s  tag=%s\n" "prod" "$active_slot" "$prod_tag"
+
+  # Stage
+  local stage_tag
+  stage_tag=$($KC_STAGE get deployment neuron-mcp \
+    -o jsonpath='{.spec.template.spec.containers[0].image}' 2>/dev/null \
+    | cut -d: -f2 || echo "unknown")
+  local stage_ready
+  stage_ready=$($KC_STAGE get pods -l app=neuron-mcp \
+    --no-headers 2>/dev/null | grep "1/1" | wc -l | tr -d ' ')
+  printf "  %-8s  tag=%-20s  pods_ready=%s\n" "stage" "$stage_tag" "$stage_ready"
+
   echo ""
-  echo "  Endpoints:"
-  echo "    prod:  https://neuron.neuralplatform.ai"
-  echo "    alpha: https://neuron-alpha.neuralplatform.ai"
-  echo "    beta:  https://neuron-beta.neuralplatform.ai"
-  echo "    gamma: https://neuron-gamma.neuralplatform.ai"
+  echo "  Prod endpoint:  https://neuron.neuralplatform.ai"
+  echo "  Prod MCP:       https://neuron.neuralplatform.ai/mcp"
 }
 
 # ── self-improvement loop (autonomous) ────────────────────────────────────────
-# When called with no args, prints the process for an autonomous agent to follow.
+# When called with no args (or 'loop'), prints the process for an autonomous agent.
 cmd_loop_process() {
   cat << 'PROCESS'
 NEURON SELF-IMPROVEMENT LOOP
 =============================
 
+Deployment model: stage (experiment) → prod (blue/green).
+
 Each iteration:
 
-1. IDENTIFY — find an improvement opportunity in the codebase.
-   Sources: Neuron backlog, observed failures, performance gaps, code review.
+1. IDENTIFY — find an improvement opportunity.
+   Sources: Neuron backlog, observed failures, graph reliability gaps,
+   session startup issues, performance, quality observations.
 
-2. IMPLEMENT — make the change on a git branch in neural-platform/neuron.
-   Commit message should describe the hypothesis being tested.
+2. IMPLEMENT — make the change on a git branch in neuron-technologies/neuron.
+   Commit message describes the hypothesis.
 
-3. CI — push the branch; CI runs lint/test/build and auto-deploys to alpha.
-   Wait for CI to complete and alpha pod to be healthy.
+3. CI — push to main; CI builds image and publishes to registry.
+   Image tag = git SHA (short). Wait for CI to complete.
 
-4. SNAPSHOT — before the next experiment touches a slot, snapshot prod DB:
-     ./scripts/neuron-self-improve.sh snapshot <slot>
-   This ensures the experiment runs against real data without touching prod.
+4. SNAPSHOT — before running the experiment, copy prod DB to stage:
+     ./scripts/neuron-self-improve.sh snapshot
+   This gives stage a real-data copy without touching prod.
 
-5. EXPERIMENT — run the experiment against the alpha endpoint:
-     https://neuron-alpha.neuralplatform.ai
-   Use real workloads. Compare against prod or other slots.
-   Multiple experiments can run simultaneously across alpha/beta/gamma.
+5. DEPLOY TO STAGE — point stage at the new image:
+     ./scripts/neuron-self-improve.sh deploy <sha>
+   Stage runs against the snapshot DB. Experiment freely.
 
-6. EVALUATE — compare results. Criteria:
-   - Does it behave correctly?
-   - Does it perform better (latency, quality, fewer errors)?
-   - Does it handle edge cases prod handles?
-   Schema changes are acceptable — they become migrations against prod on promote.
+6. EVALUATE — test the stage endpoint. Criteria:
+   - Correct behavior on same inputs?
+   - At least one dimension improved (latency, quality, error rate)?
+   - No regressions on cases prod handles?
+   - Schema changes are additive/non-destructive?
 
-7. PROMOTE WINNER — if alpha wins:
-     ./scripts/neuron-self-improve.sh promote alpha
-   This updates prod's image SHA in git → Argo CD deploys → prod is updated.
-   The old prod SHA is still in the registry — rollback is always available.
+7. PROMOTE — if stage passes, promote to prod via blue/green:
+     ./scripts/neuron-self-improve.sh promote <tag>
+   The blue-green-deploy.sh handles: scale idle slot → health check →
+   flip service selector → scale old slot to 0.
+   The old slot is kept at 0 for instant rollback.
 
-8. ROLLBACK — if something goes wrong:
-     ./scripts/neuron-self-improve.sh rollback prod <previous-sha>
+8. ROLLBACK — if something goes wrong post-promote:
+     ./scripts/neuron-self-improve.sh rollback <previous-tag>
+   Blue/green means rollback is always a slot flip — no redeployment.
 
-9. LOOP — go to step 1. Alpha is now free for the next experiment.
+9. LOOP — go to step 1. Stage is free for the next experiment.
 
 RULES
 ------
-- Never skip the snapshot step before running an experiment against a slot.
-- Prod data is never lost — slots only get copies.
-- Schema changes in experiment code → Alembic auto-migrates the snapshot DB.
-  On promote, the migration runs against prod. Always verify migrations are
-  additive/non-destructive before promoting.
-- All three experiment zones can run simultaneously with different hypotheses.
-- The registry retains every SHA — any version can be deployed to any slot.
-- This script handles git push; Argo CD handles k8s — never kubectl apply.
+- Always snapshot before deploying to stage (stage DB starts from prod).
+- Prod is never touched during the experiment — only on promote.
+- Schema changes: test Alembic migration on stage snapshot first.
+  On promote, the same migration runs against prod. Must be additive.
+- The registry retains every SHA — any version is deployable to any slot.
+- blue-green-deploy.sh handles the infra flip. This script orchestrates.
+- Never kubectl apply directly — all changes via git + Argo CD.
+  Exception: blue-green-deploy.sh uses kubectl for the slot flip because
+  that is a runtime operation, not a config change.
 
 PROCESS
 }
@@ -225,11 +217,11 @@ PROCESS
 # ── dispatch ──────────────────────────────────────────────────────────────────
 
 case "${1:-loop}" in
-  snapshot) cmd_snapshot "${@:2}" ;;
+  snapshot) cmd_snapshot ;;
   deploy)   cmd_deploy   "${@:2}" ;;
   promote)  cmd_promote  "${@:2}" ;;
   rollback) cmd_rollback "${@:2}" ;;
   status)   cmd_status ;;
   loop)     cmd_loop_process ;;
-  *)        echo "Usage: $0 snapshot|deploy|promote|rollback|status|loop" && exit 1 ;;
+  *)        echo "Usage: $0 snapshot|deploy <sha>|promote <tag>|rollback <tag>|status|loop" && exit 1 ;;
 esac