Dhaga speaks first

Your daily summary now goes out at a time you choose, read on your own clock, and a nudge reaches you before the meeting it is about instead of in the next morning's brief. Plus a search box on every table, a note you can write while you add the person, and organizations you can actually find.

For its whole life, Dhaga has been something you talk to. You forward it a message, you scan a card, you ask it a question. It answers, it files, it waits.

The one thing it never did was speak first — which is strange, because it is the only thing in your working life that knows the meeting is in ten minutes and knows the last three things you promised that person. It knew when everything was due and said nothing until the next morning's email.

That changed this week.

Your daily summary, on your clock

The daily brief — your follow-ups, your check-ins, the dates coming up — used to go out on one schedule for everybody. Whatever hour that was, it was mid-morning for somebody, breakfast for somebody else, and the night before for anyone far enough west.

There is now a card in Settings for when it lands. Pick an hour and a quarter past it, and the line underneath reads your choice back in your own time zone:

Sends at 7:15 AM in India Standard Time.

Nine in the morning is the default if you never open it. The minute picker isn't decoration, incidentally — India runs at UTC+5:30 and Nepal at +5:45, so "nine o'clock for you" was never expressible as an hour on our clock.

You also choose where it arrives: email, your linked WhatsApp or Telegram chat, or both. And because a time means nothing without a zone, the time-zone picker now sits directly above the hour it governs.

A nudge before the thing, not a summary after it

The bigger change is that Dhaga no longer waits for the morning.

It now sends a short nudge ahead of a timed follow-up or a calendar meeting — ten minutes before, which is not yet something you can change in Dhaga itself.

The part I'm most pleased with is the part you shouldn't notice. If you already told Google or Outlook to remind you thirty minutes before a particular meeting, Dhaga uses that — your setting, made once, in the calendar you already keep. Asking you to state the same preference a second time in a second app is an integration failing at its only job. Equally, a meeting you deliberately asked not to be reminded about gets no nudge from us either.

From the chat you can say "remind me in 16 minutes" and get exactly that, or snooze a nudge you have just been handed. Complete, delete or move the item and its nudge is called off. A nudge that has fallen badly late is dropped rather than delivered, because a message announcing a meeting that started twenty minutes ago is worse than no message at all.

Being honest about the plan this sits on

Proactive chat — the daily summary and the pre-event nudges arriving inside WhatsApp or Telegram — is now part of Power.

It's the first capability that separates Power from Pro. Until this week the two tiers differed only in how much AI allowance they carried, and I'd rather say plainly why this one is different: every proactive message on WhatsApp is a message we pay for, at rates that vary several times over between countries. A credit allowance can size a feature whose cost is inference. It cannot size a feature whose cost is postage.

What that means concretely:

On Free and Pro, the same daily summary and the same reminders arrive by email. Nothing you had has been taken away.
Your linked chat keeps capturing everything you forward it, on every plan, exactly as before. Replying to the bot is untouched.
The line covers Telegram too, even though sending there costs us nothing. The gate is about Dhaga starting the conversation, not about the postage — and a rule that depended on which app you happened to link would just be a rule with a loophole.
There's one further limit on WhatsApp specifically. If you've had a great many Dhaga messages in a single month, we tell you once that the rest of that month's summaries and nudges are coming by email, and WhatsApp picks up again when the month turns over.

Search that behaves like search

The People table grew a filter bar recently that could hold several companies or tags at once. Useful — and, as a customer put it, it "doesn't work at times, I have to filter twice."

They were right, and there were three separate reasons for it. All three are fixed, and while we were in there we fixed the design problem underneath them: finding one person by name on a list of people — the most ordinary thing anybody does — took four steps and two nested popups.

There is now a plain search box above every table. Type and it filters. Type Acme and it also offers you Company: Acme and Tag: Acme underneath the cursor — take one and it applies cleanly, ignore it entirely and your plain search still works, because the search never waits on the suggestions. Applied filters sit as removable chips below the box, and they live in the page's address, so a filtered view is a link you can bookmark or send to someone.

Three smaller things you'll meet sooner than you expect

Write the note while you add the person. Adding someone now has a note box on the same form. The reason you want to remember them — where you met, what they asked for — is the thing that evaporates during a redirect and a second form. It is kept word for word, and read for facts and follow-ups exactly as a note typed a minute later would be. If your month's AI credits have run out, the note is still saved: a quota should cost you suggestions, never your typing.

Type to find an event. Attaching people to an event used to mean a native dropdown of everything you'd ever run, plus a separate box for creating a new one. On Android that dropdown opened as a sheet that covered the form underneath. It's one field now: type, and it narrows on the event's name and its city, and creating a new one happens in the same box. The bulk "add these people to an event" button, which had been silently doing nothing when you typed a name rather than picking one, works.

Answer the bot in the chat. About six in ten forwarded messages come back with a "who is this about?" question, and answering one meant putting your phone down and opening the app. The bot now asks in the chat, and a number or a name answers it. The card is still raised in your inbox every time — the chat question mirrors it and never replaces it, so a question you ignore costs you nothing. It's also much less likely to ask at all: an audit of real captures found that nearly half the questions raised had already found exactly the right person and asked anyway, and a third had found nobody because a contact stored with an honorific in the name could never be matched from the first letter.

Organizations you can actually find

Team workspaces shipped a fortnight ago and were, it turned out, almost impossible to discover — the switcher sat two levels inside a menu and rendered nothing at all until you already belonged to something. It's now in the app's own navigation, on desktop and on a phone, in the same place before and after you're added to a workspace. Invitation links survive a sign-in, pending invitations are visible, and every member can see who else is in the workspace.

And an organization can now hold contacts of its own — by copy, never by move. Contributing someone writes a new record for the organization and leaves yours untouched. The two diverge from that moment: different notes, different facts, the same human. The reason is what happens when you leave. A relationship you built on your own time stays yours.

Everything copied or written into a workspace starts unpublished and shows a colleague nothing until its author publishes it. We don't call that state "private", because the organization's owner can read it — the label says not shared with colleagues, and names the owner's access in the same breath. Organizations are still set up by us on request rather than bought from the pricing page.

And the things that were broken

A short list, because they're the sort of thing you notice more than a feature:

Reminders asked for "in two minutes" could land thirteen minutes late. Delivery runs every minute now.
The map and graph detail panels blurred out the canvas they were describing, and closing the map's panel threw away the city you'd just zoomed into.
A confirmation asking which day a follow-up belonged on could get permanently stuck if you'd already dismissed the follow-up.
Re-reading a note used to delete facts and follow-ups you'd made by hand on it.
Asking Dhaga a question was charged twice.
One note could produce the same reminder twice, worded slightly differently each time.
And, less happily: for about twelve hours across the 4th and 5th of September, every forwarded WhatsApp and Telegram batch failed to save. Nothing was lost — the messages were kept and the failure recorded — but nothing was written until the fix went out the next morning. It came from a model upgrade, which is also why captures are now read by a noticeably stronger model than before.

The full list, week by week, is in the changelog.