WhatsApp and Telegram: talk to it like you text a person.

Linking your account

Linking takes two steps, one from each end, and it is built that way on purpose: a single proof is not enough to let a chat write into your workspace.

1. 1
   You declare from the app who will be writingIn Settings → Messaging. For WhatsApp, your phone number; for Telegram, a link that opens the chat with the bot already identified.
2. 2
   You message the botThe first message issues a six-digit code that arrives through the chat, never over the web. That code proves you have the phone in your hand.
3. 3
   You paste the code into the appThat closes the link: the web session proves which account it is, and the code proves which phone. It expires in fifteen minutes.

You can unlink whenever you want, and choose which workspace the bot writes to if you have more than one.

Capture: what you send, where it lands

The default behaviour is to save, not to reply. If you write something that is not a request, it saves it — because losing something you thought you saved is far more expensive than one extra message.

Forwarded audio is the least obvious of the five and it has a section of its own right below: how several of them end up in a single recap, and what comes back to you through the chat.

Forwarded audio: the summary of what people sent you

Somebody sent you three two-minute voice notes in another chat. You do not want to keep them: you want to know what they say. Forward them to the bot and what comes back is the summary.

An audio recorded in the bot's chat is a dictation, and it goes into today's note. A forwarded one is material from another conversation, and what you want out of it is the summary: that is why it becomes a recap with its transcript, and not a note. You do not have to tell it anything — WhatsApp and Telegram both flag forwarded messages, and the bot reads that flag.

1. 1
   You forward it the audioWith the chat's usual forward, one or several. The bot replies right away that it is building a recap rather than saving it into the note, with the title it took from the first sentences.
2. 2
   You have 90 seconds to add the restForwarding five audios from one thread is one act, not five: the ones arriving while the window is open join the same transcript, and every new audio slides the window another 90 seconds. You get one recap, not five. Every reply tells you how many audios it holds and how long they run: it is the only way to know you can still add another.
3. 3
   The summary comes back through the same chatWhen the window closes it is summarised once, with everything together. The summary arrives through the chat — how many audios there were, how long they ran — with the link to open the whole recap. From the last audio to the summary is a couple of minutes: the 90-second window plus whatever the summary takes.

Whatever you write along with the forwarded audio does not enter the transcript: it stays as the recap's notes. It is your context — "this is from the Acme meeting" — not something anybody said.

What comes out is a recap like any other: it carries a placeholder name with its date until the summary gives it one of its own, and from there it has all the rest — action items that turn into tasks, Atlas entities, and your workspace search.

• It is on the Pro plan, and it transcribes with the same engine as dictation: the same duration and hours caps apply.
• Audio only. A forwarded image or file follows the usual path and lands in Files.
• If the summary fails, the link still arrives: the full transcript is there in the recap.

Reminders: being told when the moment comes

Tell it "recordame mañana a las 9 llamar al contador" and at nine the alert reaches you. There is no format to learn: it understands clock times —a las 3, 15:30, 8 de la noche— and relative ones —en un minuto, en dos horas, dentro de 3 días. The bot reads Spanish, so reminders are asked for in Spanish.

What you asked for lands in three places at once, and all three matter:

You can also write them straight into the daily note, from the web or the phone, and they work the same. See Daily notes and Mobile app.

The commands

Type them or tap them: there is a drop-down menu with the same options, and tapping a row does exactly what typing it does.

On Telegram the commands autocomplete with the slash; on WhatsApp they do too, and the first time you open the chat four starter shortcuts appear. The tappable menu shows the ten most used ones — that is WhatsApp's limit — and the rest still work typed.

The ones that take no argument are commands only on their own: if you write "today I called Juan", that is a note and it is saved as one. Same with done: if it does not match any of your tasks, it goes into your daily note instead of getting lost.

How it knows whether you are talking to it or to yourself

The real problem with this channel: "tomorrow search for the Acme contract" is a note, not a search. Telling those apart well is what makes the bot usable.

1. 1
   Patterns firstKnown phrases anchored to the whole message. Instant and free, and it covers most cases.
2. 2
   Then a small classifierIf no pattern matches, a lightweight model picks one of nine labels and nothing else. It only routes: no reply you read is written by AI.
3. 3
   When in doubt, it savesThe classifier is biased toward the note. A misdetected "today" costs one extra message and you see it instantly; a misdetected capture loses something you believed you saved, and you never find out.

The assistant, and confirmations

With a leading `?` you ask the assistant about your own workspace: "where did we land with Acme?", "what did I promise Ana?". It answers with your material, not with general knowledge.

If the answer implies writing something — creating a task, changing something — the bot shows you what it is about to do and waits for a tap to confirm. Nothing is written into your workspace from chat without you saying yes.

Tasks that detection proposed can also be dismissed from chat, with a button on the message itself.

Limits and privacy

• Assistant questions have a daily cap, because they are the expensive path. Capture, commands and searches do not.
• Audio and images are on the Pro plan. Text is on every plan: send an attachment without Pro and the bot says so, and still saves whatever you wrote with it.
• Everything else respects the same plan limits as the app: messaging the bot is not a way around them.
• There are duration and size caps per attachment, and the bot tells you plainly if you go over.
• The bot only writes into your workspace: it neither reads nor touches anything from another account.
• Unlinking cuts access instantly, and the chat stops having any effect on your content.

Integrations

Both are official channels and need nothing beyond the account you already have.