diff --git a/docs/deploy-apis-flux-talos.md b/docs/deploy-apis-flux-talos.md
new file mode 100644
index 0000000..93e5ee7
--- /dev/null
+++ b/docs/deploy-apis-flux-talos.md
@@ -0,0 +1,120 @@
+# Deploying the APIs (Flux GitOps → Talos)
+
+> Handoff doc. The APIs are **not** deployed by Gitea Actions — they are deployed
+> by **Flux** running on the Talos Kubernetes cluster. The Gitea Actions in this
+> repo only build the docs-site image, validate Flux manifests, and publish
+> OpenAPI specs to RapidAPI.
+
+## How a deploy actually happens
+
+```
+API repo (e.g. leeworks-agents/zip-enrichment)
+  └─ its own CI builds & pushes  registry.leeworks.dev/zip-enrichment/server:<tag>
+       └─ Flux image-automation (flux/image-automation/) rewrites the
+          {"$imagepolicy": "flux-system:<api>"} marker in flux/<api>/helmrelease.yaml
+            └─ Flux GitRepository "api-company" (polls main every 5m)
+                 └─ HelmRelease per API (flux/<api>/helmrelease.yaml, bedag/raw chart)
+                      └─ Deployment rolls out in the cluster namespace
+```
+
+Each API has its own directory under `flux/`:
+
+| API | Namespace | HelmRelease path | Image |
+|---|---|---|---|
+| zip-enrichment | `zip-enrichment` | `flux/zip-enrichment/helmrelease.yaml` | `registry.leeworks.dev/zip-enrichment/server` |
+| holidays | `holidays` | `flux/holidays/helmrelease.yaml` | `registry.leeworks.dev/holidays/server` |
+| air-quality | `air-quality` | `flux/air-quality/helmrelease.yaml` | `registry.leeworks.dev/air-quality/server` |
+| vin-decoder | `vin-decoder` | `flux/vin-decoder/helmrelease.yaml` | `registry.leeworks.dev/vin-decoder/server` |
+
+(Confirm each path's exact image with `grep -r imagepolicy flux/`.)
+
+## Prerequisite: the manifests must be live in the cluster's Flux source
+
+`flux/api-company-source/gitrepository.yaml` is **reference only**. The
+*authoritative* copy must be committed to **`0xWheatyz/Talos`** at:
+
+```
+testing1/first-cluster/cluster/flux/api-company/
+```
+
+If that path does not point Flux at this repo's `flux/` directory, Flux never
+sees these HelmReleases and nothing deploys. Verify the Talos repo references
+this repo's `main` branch and that a Flux `Kustomization` includes the
+`api-company` path.
+
+Per-API the cluster also needs (already templated under `flux/<api>/`):
+- `namespace.yaml` — the target namespace
+- the `gitea-registry` imagePullSecret in that namespace
+- `externalsecret.yaml` — pulls API keys (e.g. RapidAPI) via external-secrets
+- `servicemonitor.yaml` — Prometheus scraping (optional for deploy)
+
+## One-time local setup (machine with cluster access)
+
+You need tools that are **not** installed on the dev machine yet:
+
+```bash
+# Talos kubeconfig — export from the Talos controlplane, e.g.:
+#   talosctl kubeconfig ~/.kube/talos-leeworks
+export KUBECONFIG=~/.kube/talos-leeworks
+kubectl cluster-info        # must succeed before continuing
+
+# Flux CLI
+brew install fluxcd/tap/flux
+# Helm (optional, for debugging charts)
+brew install helm
+
+flux check                  # confirm Flux is installed & healthy in-cluster
+```
+
+## Deploy / sync all APIs
+
+```bash
+export KUBECONFIG=~/.kube/talos-leeworks
+
+# 1. Pull the latest main into the cluster's Git source
+flux reconcile source git api-company -n flux-system
+
+# 2. Apply the manifests (name may differ — check: flux get kustomizations -A)
+flux reconcile kustomization api-company -n flux-system
+
+# 3. Reconcile each API's HelmRelease
+flux reconcile helmrelease zip-enrichment -n zip-enrichment
+flux reconcile helmrelease holidays       -n holidays
+flux reconcile helmrelease air-quality    -n air-quality
+flux reconcile helmrelease vin-decoder    -n vin-decoder
+
+# 4. Verify everything is Ready
+flux get helmreleases -A
+kubectl get pods -A | grep -E 'zip-enrichment|holidays|air-quality|vin-decoder'
+```
+
+## Troubleshooting
+
+```bash
+# Why is a release not Ready?
+flux get helmrelease <name> -n <ns>
+kubectl describe helmrelease <name> -n <ns>
+
+# Is image automation picking up new tags?
+flux get image policy -n flux-system
+flux get image update -A
+
+# Pod won't start (image pull / secret issues)
+kubectl describe pod <pod> -n <ns>
+kubectl get events -n <ns> --sort-by=.lastTimestamp | tail -20
+```
+
+## Force a fresh deploy of a single API
+
+```bash
+flux suspend helmrelease <name> -n <ns>
+flux resume  helmrelease <name> -n <ns>   # triggers a fresh reconcile
+# or restart the workload directly:
+kubectl rollout restart deployment/<name> -n <ns>
+```
+
+## Related docs in this repo
+- `docs/operator-runbook.md` — day-2 operations
+- `docs/registry.md` — container registry (`registry.leeworks.dev`) setup
+- `docs/secrets-checklist.md` — required cluster secrets
+- `flux/image-automation/` — automatic image tag bumping
diff --git a/scripts/setup-sibling-repos-token.sh b/scripts/setup-sibling-repos-token.sh
new file mode 100755
index 0000000..5aadd94
--- /dev/null
+++ b/scripts/setup-sibling-repos-token.sh
@@ -0,0 +1,55 @@
+#!/usr/bin/env bash
+# Create a Gitea PAT with read access to the sibling API repos and store it as
+# the SIBLING_REPOS_TOKEN action secret on leeworks-agents/api-company.
+#
+# Why this script exists:
+#   - Gitea's build-docs workflow checks out sibling repos. The auto-injected
+#     GITEA_TOKEN is scoped to THIS repo only and cannot read them, and
+#     GITEA_TOKEN is a reserved secret name that cannot be overridden.
+#   - `tea` cannot CREATE a PAT (no such command), and Gitea's token-creation
+#     API requires BASIC AUTH (your password) — a token cannot mint a token.
+#   - `tea` CAN set the action secret using its existing login.
+#
+# So: this prompts for your password ONCE, mints the PAT via the API, and pipes
+# it straight into `tea` as the secret. The PAT value is never written to disk.
+#
+# Usage:  bash scripts/setup-sibling-repos-token.sh
+set -euo pipefail
+
+GITEA_URL="https://gitea.leeworks.dev"
+GITEA_USER="0xWheatyz"
+REPO="leeworks-agents/api-company"
+SECRET_NAME="SIBLING_REPOS_TOKEN"
+TOKEN_NAME="sibling-repos-readonly-$(date +%Y%m%d)"
+
+command -v curl >/dev/null || { echo "curl required"; exit 1; }
+command -v tea  >/dev/null || { echo "tea required";  exit 1; }
+command -v python3 >/dev/null || { echo "python3 required"; exit 1; }
+
+echo "Gitea user: $GITEA_USER  ($GITEA_URL)"
+read -r -s -p "Gitea password (for $GITEA_USER): " GITEA_PASS
+echo
+
+# Create a read-only PAT. scope read:repository lets the build-docs workflow
+# clone the sibling repos. Adjust scopes here if your Gitea version differs.
+resp="$(curl -fsS -X POST \
+  -u "${GITEA_USER}:${GITEA_PASS}" \
+  -H 'Content-Type: application/json' \
+  -d "{\"name\":\"${TOKEN_NAME}\",\"scopes\":[\"read:repository\"]}" \
+  "${GITEA_URL}/api/v1/users/${GITEA_USER}/tokens")" || {
+    echo "Token creation failed. Check password / 2FA (2FA blocks basic-auth token creation)." >&2
+    exit 1
+  }
+unset GITEA_PASS
+
+PAT="$(printf '%s' "$resp" | python3 -c 'import sys,json; print(json.load(sys.stdin)["sha1"])')"
+[ -n "$PAT" ] || { echo "Could not parse token from response: $resp" >&2; exit 1; }
+echo "PAT '${TOKEN_NAME}' created."
+
+# Store it as the action secret via tea (overwrites if it already exists).
+printf '%s' "$PAT" | tea actions secrets create "$SECRET_NAME" --repo "$REPO" --stdin
+unset PAT
+
+echo "Secret '${SECRET_NAME}' set on ${REPO}."
+echo "Verify:  tea actions secrets list --repo ${REPO}"
+echo "Then re-run build-docs:  tea actions runs ...  (or push to main / workflow_dispatch)"