Everything genuinely open right now. Last updated 18 Aug 2026.
Things only you can do — a click, a credential, or a check.
USE_REAL_DATAFORSEO is now true in production (was false all session) — so activating "ORDER_12 LLM Visibility Probe" (weekly Monday 8 AM IST) means real spend from that point on, roughly $0.40-0.50/week at current prompt/model count. The other five ("ORDER_09", "ORDER_07", "ORDER_10", "ORDER_11" daily 8 AM IST, "ORDER_13" weekly Monday 9 AM IST) are free or DataForSEO-metered-but-cheap. All six are imported but inactive — n8n's CLI can't toggle Active outside queue mode, so this is a one-time manual step per workflow.
Open http://187.127.153.207:5678/home/workflows → open each → toggle Active (top right).
NotifyService is log-only until you provide a webhook URL or bot token — nothing was fabricated here on purpose.
Send the webhook URL / bot token whenever it's ready and it can be wired into NOTIFY_WEBHOOK_URL the same session.
Confirmed via DataForSEO: the listing is genuinely claimed (is_claimed=true), so this is worth doing — but the whole setup (Google Cloud project, OAuth client, the interactive consent flow logging in as the account that owns the listing) needs a browser and your Google login. No part of it can be automated from here. A claimed earlier "pending application, support case" was checked and found to be untraceable on the real infrastructure — this starts from zero. Verified against Google's own docs 18 Aug 2026: the reviews endpoint specifically sits behind a separate "Basic API Access" request form with a real 7-10 business day review window and a zero-quota project until approved — submit that first, before anything else here.
In order: (1) submit Google's Business Profile API access request form for the reviews scope and wait — this is the real bottleneck, budget over a week; (2) meanwhile create the Cloud project + OAuth client (Desktop app type) → CLIENT_ID + CLIENT_SECRET; (3) once approved, run the one-time consent flow as the listing-owner account → REFRESH_TOKEN; (4) call accounts.list / locations.list → ACCOUNT_ID + LOCATION_ID. Send all 5 values whenever they're ready and ORDER_11 gets wired to them the same session — built against the real thing, not guessed. Until then ORDER_11 stays on mock and its n8n schedule shouldn't be activated for real GBP data.
Optional. Say the word on any of these whenever you want it.
The core pages work well. If anything specific starts to bother you in daily use, that's the moment to ask for it rather than guessing at polish now.
Practo/Lybrate/1mg/Sulekha are now verified and live in sameAs. The one remaining schema gap is a logo image URL — the site has no logo asset at all (text link, not an image), so this stays omitted rather than guessed.
Whenever a real logo image exists somewhere, send the URL and it gets added the same session.
Not urgent, not blocking anything — just worth knowing about.
Ubuntu flagged a newer kernel available while installing the Postgres client. Not urgent, but it needs a reboot to apply — worth scheduling during a genuinely quiet window since it will briefly take down every container on the box, including the other Hermes system, not just this one.
Fixed the traefik crash-loop (it was never actually broken-by-itself — it just could never win the port-80/443 race against nginx). n8n, waha, openwa, searxng, and the Hermes Agent terminal are now reachable at their *.srv1680582.hstgr.cloud URLs for the first time, routed through nginx directly, with the agent terminal's password protection re-verified intact. HTTPS certs failed with a genuine external cause: these subdomains have IPv6 records, and Let's Encrypt's validators can't reach this VPS over IPv6 — not fixable from inside the VPS.
Either file a Hostinger support ticket about IPv6 routing for this VPS, or provide Hostinger DNS API credentials so certs can be issued via DNS-01 challenge instead (sidesteps IPv6 entirely).
Tested live (real ChatGPT call, real web search, 18 Aug 2026): asking "best cardiologist in Yelahanka" returned three competitors (two Sparsh Hospital doctors, one Manipal Hospitals Yelahanka) and no mention of Samiksha. DataForSEO's own mention index shows zero prior ChatGPT citations of the domain too. This is exactly the gap ORDER_12 now tracks weekly — not a guess, a measured baseline.
Reviewers Anjali, Deepa, Kavita, Priya, and Ramesh are still sitting in the GBP review draft queue — genuinely unclear whether these are real patient reviews from an earlier session or test data, so they were deliberately left alone rather than guessed at during today's queue cleanup (101 confirmed-test content drafts and 7 confirmed-test GBP reviews were removed).
If you recognize these as real reviews, approve/edit as normal. If they're test data, they're safe to delete from /gbp-reviews.
Now that USE_REAL_DATAFORSEO is on, tested directly: the reviews endpoint returns a real 404 on this account. A working sibling endpoint (my_business_info) gives rating/is_claimed/hours but not individual review text, so it can't substitute. ORDER_11 safely falls back to mock every time rather than failing — nothing is broken, it just never pulls real reviews yet. Fixing this needs DataForSEO's actual docs or reading CarePulse's own working sync-gbp-reviews.mjs, neither available here.
Checked live via DataForSEO (18 Aug 2026): heartanddiabeticcare.com has only 6 backlinks from 6 referring domains, with an overall spam score of 64/100. The domains themselves look off-topic for a cardiology clinic — a US real-estate site (homesforsaleoldgreenwichct.com), a generic "us.com" domain, a site named betulcrime.com. None look like the legitimate healthcare citations (Practo, Justdial, 1mg, GBP) this project's own docs have long recommended.
Worth a manual look at where these links actually came from, and considering a disavow file for the spammiest ones — ORDER_13 only measures, it never acts automatically.