The daily rhythm: inbox, meetings, and the evening call

The daily rhythm: inbox, meetings, and the evening call

Station began as a place to think — capture, challenge, plan, do. The past few months have been about something different: making Station show up on its own schedule, not just when I open a browser tab. It now bookends my working day and fills the gaps in between.

● As at 5 September 2026 · times are Australian Eastern Standard Time

← Back to the Station overview · See also: the grand tour · capturing things · from idea to done


The morning — three things, and nothing else shouting

The front page opens on a short answer to "what actually matters today": up to three things I’ve ticked from my open work, whatever I’ve pinned there myself, and counts of what’s waiting. Tick a fourth and the rest go grey — that’s the cap, not a bug.

The three aren’t copies of anything. A priority is the task, so finishing it here finishes it on its plan too. At the end of the day Station offers to carry the unfinished ones into the next working day, pre-ticked; anything I untick simply stays on its plan.

There are two ways to make the morning prompt go away and they mean different things: snooze quiets the prompt for half an hour and leaves my picks alone, while dismiss for today says there are no priorities today — for a day off.

Station's front page in September 2026: a worklist showing today's top three picked tasks, a row of pinned items, clickable counts of what is waiting, and columns for what arrived recently and what is in flight.
Station’s front page in September 2026: a worklist showing today’s top three picked tasks, a row of pinned items, clickable counts of what is waiting, and columns for what arrived recently and what is in flight.

Inbox triage — the mail gets read before I read it

Station watches exactly two mailboxes: my personal one and my Microsoft MVP one. My work mail is not one of them and never has been — that stays where it belongs. What does land, Station classifies with a model that runs on my own hardware, and it learns from my corrections rather than sending anything to a cloud service to be read.

The rule I care most about: only two actions are ever taken automatically — delete and decline. Everything else (reply, research, turn into a task) is surfaced for me to press. And before an automatic delete or decline fires, Station judges whether the message is the routine version of the thing it thinks it is; anything unusual is downgraded to a suggestion and left in the inbox. Polling either mailbox never changes it.

Every row shows a confidence — ● high · ◐ medium · ○ low — and the word beside it says where the suggestion came from: a rule I wrote, the local model, or a school-portal match. Low confidence isn’t a warning about the email; it’s Station telling me it’s guessing.

Three things arrived since July that make it much less work:

  • "Always do this with mail like this." Rather than waiting for Station to work something out, I open any email and it fills in a rule from that message — the whole sender domain or just that address, optionally a word from the subject, and what to do. It takes effect at the next check. A rule is also the safer of the two routes, because the routine check still applies.
  • It learns one bucket per family, not one per email. Teaching it about a newsletter used to teach it about that newsletter. Since 29 August 2026 a correction lands against the whole family of similar mail.
  • Acknowledge and Dismiss are different. Both clear the item. Acknowledge teaches Station nothing, so I’ll be shown the next one like it. Dismiss also counts against ever automating that kind of item. Use Dismiss when you want less of something.

Everything Station did on its own is listed under Done for you, and every rule I wrote can be changed or removed on its own page — the screen below. Nothing happens invisibly.

The Rules and exclusions page: the standing rules I have written for incoming mail — sender or domain, an optional subject word, and the action to take.
The Rules and exclusions page: the standing rules I have written for incoming mail — sender or domain, an optional subject word, and the action to take.

Meetings — the commitments I make out loud

The most dangerous place for a commitment is a meeting. I say "I’ll send you that report" and it evaporates before the call ends.

Meeting audio goes through Whisper, locally, and Station reads the transcript looking for one thing only: first-person, future-tense commitments that I made. Not everyone’s action items — mine. Each candidate comes up for a quick confirm or reject, and every decision trains the extractor. I can correct the transcript, and keep my own vocabulary list for the names and jargon Whisper trips over.

The big change landed on 28 August 2026: a recording now knows which meeting it came from. Station keeps a copy of my work calendar, reads the attendee domains out of the invitation, and files the recording against the project those domains belong to. The meetings list is headed by the meeting’s actual name rather than a timestamp, badged internal, external or unknown with the domains that decided it, and two one-click corrections are available that the matcher will never overwrite.

A recording that couldn’t find a meeting isn’t abandoned either — it gets looked at again once the calendar catches up.

And because the transcripts are there anyway: there’s a swear jar. A persistent, per-word tally of what I said on which call. It survives the recording being deleted, which feels like a design decision aimed directly at me.

(No screenshot of the meetings list. It is, by definition, a list of the people I met — and three of their names came through the redaction pass intact when I tried, even on the sanitised instance. Description only, until that’s fixed.)


The evening call — Station rings me, in my voice

At the end of every weekday my phone rings. Full-screen incoming call, answer or decline, the whole thing. It’s Station, speaking in a clone of my own voice, asking how the day went.

It’s a real interview, not a form. It knows what was on my plate, asks what moved, and pulls the loose threads. Out of that come:

  • Journal entries, written per workspace, so client notes land in the client’s workspace and homelab notes in the homelab one.
  • Untracked work, captured — the things I did but never put into Station become tasks that are created and closed in one motion, after I approve them out loud. The record stays honest without me doing data entry.

Four things about it got noticeably better over August, and they’re all about not being annoying:

  • It stops after about two seconds rather than reading my whole calendar out. The original version was a monologue; now it’s a conversation.
  • It doesn’t read personal appointments aloud any more. That one was a privacy fix, and an obvious one in hindsight.
  • The phone shows how far through the debrief you are, so it isn’t an open-ended commitment at 6pm.
  • Snoozing it actually brings it back. It used to just… not.

Talking to yourself was supposed to be a warning sign. Turns out it’s a productivity system.

The Debriefs review page: an evening call session with its transcript, the journal entries it wrote, and the tasks it proposed for approval.
The Debriefs review page: an evening call session with its transcript, the journal entries it wrote, and the tasks it proposed for approval.

Journals — a diary I don’t have to write

Every workspace has its own journal. The debrief writes them nightly, and anything notable from the day’s activity lands there too. Months from now, "what actually happened on that engagement?" has a real answer, written the same evening rather than reconstructed from a calendar.


The follow-up loop — features don’t get to rot

This closes a loop most tools never acknowledge: the gap between shipping something and knowing whether it earned its place.

Every feature that ships is registered with a plain-language test plan, and Station brings it back at 2, 7, 14 and 30 days — one escalating ladder, not four separate nags. Did you actually use this? Is it doing what you wanted? The check-ins are grounded in a local, append-only usage log — nobody’s analytics platform — drip-fed one a day so it never becomes homework. "Not now" reschedules, "snooze" waits a week, and "don’t ask again" hides that feature’s remaining check-ins without deleting its history.

The page has grown two jobs beyond that:

  • Needs a look appears at the top only when a background job or a knowledge source keeps failing — each row carrying ◐ warn as a shape and a word, a one-line reason, and a link to where it can be fixed. When there’s nothing wrong, the block is completely absent. Frequent jobs need three consecutive failures before they’ll bother me; a job that runs weekly can speak up after one.
  • "Station learned this — is it right?" shows the newest memory facts waiting for approval, written as ordinary sentences, with ● Accept and ○ Reject. Nothing enters Station’s memory about me without that second step.

Two honesty fixes landed on 3 September 2026 that I’m oddly fond of. Anonymous probes hitting the login page were being counted as me using a feature — they aren’t any more. And a feature with genuinely no use across a full 60-day window now has its check-ins parked rather than asked again forever. The answer to "did you use it?" was already in, forty times.

The house rule underneath all this: not registered = not shipped.

The Follow-ups page: today's check-in on a shipped feature with its test-plan steps, a Needs a look block flagging a failing job with a warn shape and the word warn, and a block of memory facts awaiting Accept or Reject.
The Follow-ups page: today’s check-in on a shipped feature with its test-plan steps, a Needs a look block flagging a failing job with a warn shape and the word warn, and a block of memory facts awaiting Accept or Reject.

The shape of it

Morning — the three things, the daily email, and at most one push notification, sent only when there’s something I can actually act on. During the day — inbox triage and meeting capture feed the queue quietly. Evening — the phone rings, the day gets debriefed, the journals write themselves, and Station is allowed to say "the day is done". Overnight — Hermes reads the day’s thinking and has something to say in the morning. That’s the piece that was "next" the last time I wrote this page; it shipped on 1 August 2026, and it’s suggest-only — it can look, not touch.

And underneath all of it, the follow-up loop keeps the whole system honest about itself.

Every agent that runs on its own, what it’s allowed to touch and what it reports: Helpers and agents →.


← Back to the Station overview.