Daily digest in Topologix

Monitoring & Observability

The daily digest is an e-mail an organisation can turn on under Organisation → Settings → Daily digest. Once a day, at the hour the owner picks in the organisation's time zone, Topologix computes the numbers itself (a facts pack: open problems, alert groups, silent sources, hardware findings, changes, log-error types, service-call edges) and a rule-based list of candidates, then hands both to the organisation's Assistant AI model — the one the monitoring bot uses — which investigates with read-only tools and writes two sections: What changed and Suspicious — investigate. The numbers are never written by the model.

Requirements

• An AI model that serves the organisation's Assistant role (Organisation → AI models): the organisation's own provider, or the platform Claude key where the organisation may use it. Without one, the digest cannot be turned on.
• SMTP configured by a super admin (System settings → Email). Without it nothing is sent and the card shows the error.
• The organisation owner (or a super admin) to turn it on and keep the recipient list; admins see the card read-only and can preview it or send themselves a test.

Turn it on

1. Op
2. Pi
3. Ad
4. Ti

What the e-mail contains

Sections

Every digest has the same shape, also on a quiet day (a short “all quiet” mail):

• Status now — first the incidents (measured impact on critical / important services): yesterday's with how long each lasted and its suspected cause ("2 incidents yesterday — checkout throughput −35 % for 41 min (cause: cascade from esxi142), Online players −18 % for 12 min (no cause detected)") and the ones open now, each linked to the incident; then open problems by severity, alert groups by class ("3 alert groups — 1 cascade, 1 common cause, 1 burst", each with its title and why line), stale latches, silent sources, hypervisor findings and services flagged unusual. With incidents the subject says so: "2 incidents yesterday (41 min, 12 min) — 1 open now".
• Yesterday vs the day before vs the same day last week — a three-column table: incidents, incident outages, the median incident duration and time to acknowledge (once there has been one), problems opened and resolved, median time to resolve, flapping subjects, hardware findings, kernel crashes, changes (failed or rolled back too), log-error volume and new error types, service-call edges whose errors or volume moved. A column older than the stored history says “no data yet”.
• What changed and Suspicious — investigate — written by the AI model; each suspicious item says why, what was checked, the next step and links back into Topologix (at most 12 items).
• A footer: the AI status line (which model, how many actions), “numbers as of”, “covers what the organisation owner can see”, and the unsubscribe link.

When the AI is not available

If the model fails, times out, answers in the wrong format or the AI pack is used up, the digest still goes out with its numbers and the rule-based suspicious list, under the banner “AI investigation unavailable today — <reason>”. The settings card shows the reason on its status line.

Privacy

The digest covers what the organisation owner can see: private flows of other members are left out. The whole e-mail passes through the organisation's data-redaction rules; no credentials and no raw log lines go into the facts the model reads.

Recipients, opt-in and unsubscribe

Who receives it

The owner's list (≤ 25 addresses) plus every active member who opted in from their account menu → Alert notifications → Daily e-mail digest (≤ 100 members per run). An address that is both receives one mail. Each recipient gets their own mail — nobody sees the others' addresses.

Unsubscribe

Every mail carries a one-click unsubscribe link (and the List-Unsubscribe headers mail clients show as a button). A listed address that unsubscribes is skipped from then on — even if the owner adds it again; the card shows it greyed, “unsubscribed <date>”, and only the owner can Clear it (for when the person asks to receive it again). A member who unsubscribes is simply opted out.

Timing and cost

When it is sent

At the chosen hour in the chosen zone, within about a minute. “Yesterday” and “the same day last week” are calendar days in that zone. If the server is down at that hour the digest still goes out within the next 6 hours; after that the day is skipped (“missed the slot”) rather than sent late. Each day is sent once — a restart never sends it twice — and daylight-saving changes neither skip nor repeat a day.

AI cost

Each digest uses about 3–7 AI actions (one per request to the model; at most 6 rounds in 4 minutes). On the platform key they count against the organisation's AI pack; through the organisation's own provider they are counted but never limited. The preview never uses the AI; Send me a test with AI does.

Where admins see it

The card's status line: “Last sent <time> to N recipients · AI: <model> · N actions”, or the reason, and the last error in red. Super admins see every organisation's digest (on/off, last sent, AI status, last error) in the Daily digest column of Admin → Organisations.

Good to know

• The digest is built on the primary server process by a 60-second worker (at most 2 organisations per pass; log line [digest] db-cost: sweep …).
• Only the last rendered body is kept (for the admin side); the e-mails themselves are not archived.