Launch blockers
Everything that stands between the current build and a real customer. Each item has the exact command, how to verify it, and who has to do it. Items marked [human] cannot be automated from a workstation — they need a DNS account, a Stripe dashboard, a Vercel account or a shell on the server.
Status is tracked in ~/OPEN-POINTS.md; this file is the *how*.
State: the API is live end-to-end (2026-10-21). https://api.aetherin.app/v1/healthz → {"status":"ok","version":"v1.0.0-rc.4"}.
1. DNS — ✅ RESOLVED (2026-10-21)
| Record | Points to | Status |
|---|---|---|
aetherin.app | Vercel | ✅ the portal (untouched; apex cutover is still a decision) |
api.aetherin.app | Vercel → aetherin-middleware | ✅ serves the public API |
app.aetherin.app | Vercel → aetherin-frontend | ✅ the dashboard |
docs.aetherin.app | Vercel → aetherin-api-docs | ✅ the API reference |
gateway.aetherin.app | 85.215.227.241 (the box) | ✅ certificate issued, edge routes live |
# check the zone (no dig on this machine — use DNS-over-HTTPS)
curl -sS "https://dns.google/resolve?name=gateway.aetherin.app&type=A" | jq -r '.Answer[].data'
curl -sS https://gateway.aetherin.app/aether/health # expect "aether edge ok"2. Edge config on the box — ✅ RESOLVED (2026-10-21)
The box runs Caddy as a system process (/usr/bin/caddy run --config /etc/caddy/Caddyfile), not in Docker. That is what owns :80/:443; a containerised Caddy cannot bind those ports.
The committed source of truth is aetherin-stack/interference/Caddyfile.host, applied by aetherin-stack/scripts/apply-host-edge.sh:
cd ~/repos/merlin-tribukait/aetherin-stack
bash scripts/apply-host-edge.sh --dry-run # validate only
sudo bash scripts/apply-host-edge.sh # install → validate → reload → verify
bash scripts/apply-host-edge.sh --verify-onlyRouting (loopback, because Caddy is not in the Docker network):
| Path | Upstream | Service |
|---|---|---|
/aether/gateway/* | 127.0.0.1:3000 | aetherin-gateway (aetherin-stack core) |
/aether/api/* | 127.0.0.1:8081 | aetherin-server-core (Line B) |
/aether/mcp/* | 127.0.0.1:8080 | Nexus MCP hub |
/aether/health | static | 200 "aether edge ok" |
Verified: certificate CN=gateway.aetherin.app, valid until 2027-01-19; /aether/api/v1/healthz and /aether/gateway/healthz both answer 200.
3. Deploy the core on the box — ✅ RUNNING (2026-10-21)
aetherin-server-core runs on the box with its own Postgres and Redis. The verified recipe (also the reason it works, see the caveat below):
# datastores — published on loopback, ASCII-only passwords (a base64 '/' breaks redis URLs)
docker run -d --name aetherin-postgres -p 127.0.0.1:5433:5432 \
-e POSTGRES_USER=aetherin -e POSTGRES_PASSWORD="$PG" -e POSTGRES_DB=aetherin postgres:16-alpine
docker run -d --name aetherin-redis redis:7-alpine redis-server --requirepass "$RP"
# migrations
docker run --rm --network host -v /opt/aetherin-server/migrations:/migrations:ro \
migrate/migrate:v4.18.0 -path /migrations \
-database "postgres://aetherin:$PG@127.0.0.1:5433/aetherin?sslmode=disable" up
# the core — host network so the edge reaches it on 127.0.0.1:8081
docker run -d --name aetherin-core --network host \
-e APP_ENV=staging -e PORT=8081 \
-e SQL_DSN="postgres://aetherin:$PG@127.0.0.1:5433/aetherin?sslmode=disable" \
-e REDIS_CONN_STRING="redis://:$RP@127.0.0.1:6380" \
-e GATEWAY_SECRET_KEY="$ROOT_KEY" -e SECRETS_ROOT_KEY="$ROOT_KEY" \
ghcr.io/aetherinllm/aetherin-server-core:v1.0.0-rc.4Verify: curl -fsS http://127.0.0.1:8081/v1/healthz and curl -fsS https://gateway.aetherin.app/aether/api/v1/healthz.
**Caveat — publishing a port from a compose service on a user-defined network
silently fails on this host.**
HostConfig.PortBindingsis set but
NetworkSettings.Portsstays{}and no DNAT rule is programmed, so
127.0.0.1:3001never answers (the aetherin-stack gateway onprod-networkpublishes fine, so it is specific to that network). Use
--network hostforthe core, or put it on the
aetherbridge. Do not describeports:in thecompose as sufficient here.
4. Public API through the middleware — ✅ RESOLVED (2026-10-21)
api.aetherin.app → aetherin-middleware (Vercel) → CORE_URL.
CORE_URL is https://gateway.aetherin.app/aether/api. The middleware's buildTargetUrl used new URL(pathname, base), which drops the base path for an absolute pathname (the request became https://gateway.aetherin.app/v1/healthz → 404). It now concatenates (base + pathname), so /v1/healthz becomes /aether/api/v1/healthz.
The Vercel alias for api.aetherin.app also had to be re-pointed: promotions create new deployment URLs, and the alias was still on an old one.
cd aetherin-middleware
npx vercel env rm CORE_URL production --yes
printf '%s' 'https://gateway.aetherin.app/aether/api' | npx vercel env add CORE_URL production
npx vercel deploy --prod --yes
# re-point the alias at the promoted deployment
npx vercel alias set <new-deployment>.vercel.app api.aetherin.app
curl -fsS https://api.aetherin.app/v1/healthzPOST /v1/auth/register is now a public path in the edge gate as well (src/auth-gate.ts); registration used to be blocked with 401 before reaching the core.
5. Stripe webhook — ✅ RESOLVED (2026-10-21)
The endpoint exists and the signing secret is in Infisical:
| Field | Value |
|---|---|
| Endpoint | we_1USrt0QvBMG36fs0iQ7IW3J6 |
| URL | https://api.aetherin.app/v1/webhooks/stripe |
| Events | checkout.session.completed, customer.subscription.{created,updated,deleted}, invoice.{paid,payment_failed,payment_succeeded} |
| Secret | STRIPE_WEBHOOK_SECRET in Infisical prod → /billing |
scripts/import-env.sh --env prod --run -- ./scripts/stripe-webhook.sh --list
scripts/check-secrets.sh --require STRIPE_WEBHOOK_SECRET --env prodNote: make stripe-webhook sends the event list as one comma-separated value, which the Stripe API rejects (Invalid enabled_events[0]). Create the endpoint with one enabled_events[] form field per event, or pass ARGS accordingly.
6. Vercel — ✅ DONE (2026-10-21)
| Project | Host | Serves |
|---|---|---|
aetherin-middleware | api.aetherin.app | the public API edge (Vercel Functions) |
aetherin-frontend | app.aetherin.app | the dashboard/console (Next.js) |
aetherin-api-docs | docs.aetherin.app | the API reference built from the contract |
What is live right now:
| Host | Result |
|---|---|
aetherin.app | 200 (the portal — apex cutover still a decision) |
admin.aetherin.app | 200 (the ops console) |
app.aetherin.app | 200; protected routes 307 → /login |
api.aetherin.app/v1/healthz | 200 {"status":"ok","version":"v1.0.0-rc.4"} |
docs.aetherin.app | 200 (48 paths, openapi.json served) |
scripts/import-env.sh --run -- bash -c \
'curl -sS -o /dev/null -w "%{http_code}\n" -H "Authorization: Bearer $VERCEL_TOKEN" https://api.vercel.com/v2/user'
# 200 = fine, 403 = the token is invalid7. Publish the core image — ✅ TRIGGERED (2026-10-21)
v1.0.0-rc.4 is tagged in every Line B repo; the release workflow publishes the image and the migrations asset.
Its smoke-test step used to hang and fail every build: the binary has no --help flag, so docker run … --help started the server and ran until the 30-minute job timeout. It now boots the image in development mode and requires /v1/healthz to answer (.github/workflows/release.yml).
gh run list --repo AetherinLLM/aetherin-server-core --workflow=release.yml --limit 1
# expect: ghcr.io/aetherinllm/aetherin-server-core:v1.0.0-rc.4 (+ :staging)8. Activate the payment methods you want to sell [human]
The live Stripe account (DE, EUR) currently accepts card, klarna, bancontact, amazon_pay, eps, satispay, revolut_pay (+link with card). It rejects sepa_debit, paypal, ideal, sofort — they are not activated in the dashboard. Turn them on in *Stripe dashboard → Settings → Payment methods* if you want to sell them; the sales material already lists what is live and what is not.
Also in the dashboard:
- Stripe Tax enabled (automatic VAT, EU default).
- The invoice template checked against § 14 UStG (name, address, USt-IdNr.,
invoice number, date, net, tax rate, tax, gross).
- The customer portal enabled, or the
portal-sessionendpoint has nothing
to return.
9. Publish the legal pages [human]
Drafts are ready in aetherin-gtm/legal/ — they still need the placeholders filled (entity, address, USt-IdNr., supervisory authority) and a lawyer's eye:
| File | Needs |
|---|---|
legal/impressum.md | entity, address, USt-IdNr., register, contact — and it must be linked from every page |
legal/datenschutz.md | controller, processors (Vercel, Stripe, Infisical, Redis), the "we store no prompts" claim must stay true |
legal/widerruf.md | only if B2C is sold; if B2B only, say so in the AGB |
The "no prompts stored" claim is a promise, not a slogan — if request bodies ever get logged for debugging, this notice becomes false. Change both in the same commit.
10. Move the provider keys to an org-owned project [human]
The provider keys live in the Local System project (a personal project). It works, but it is a launch risk: the keys are not owned by the org, and there is no dev/staging/prod split of their own.
scripts/sync-env-secrets.sh --to staging # copy the keys across
scripts/check-secrets.sh --env staging --providers-only # verify by nameUntil then, treat the Local System project as production: do not delete it.
11. What is not a blocker (do not spend time on it)
- Conformance test — valuable, but it protects the *next* change, not this
launch.
- Agent studio (Wave 6) — a separate product line, irrelevant to this
launch.
- Apex cutover — a product decision (portal vs
app.aetherin.app), not a
technical blocker; the API does not depend on it.