The assistant you talk to in the app is reachable from your phone over WhatsApp or Telegram. It is the same conversation and the same assistant — not a notification feed and not a second, weaker bot. It can walk you through your work, answer questions about it, and carry out the things the page can carry out.

Settings → Phone & chats#

Every chat that reaches Taskuary from outside is set on one page, Settings → Phone & chats. The WhatsApp and Telegram connection cards link to it.

• The Assistant from your phone — your private Assistant chat on each channel: which chat it is (or the steps to pick one), whether it listens any time or only during a walk, the morning line, and Walk me through them in WhatsApp (or Telegram) to hand the walk over from the desktop.
• Agent channels — shown only with Developer mode on: every chat bound to a coding agent, what it talks to, its repository and its people, with Unbind, and Add a channel (below).

The doorways#

A doorway is one chat, on one channel, that the assistant is allowed to speak in. Set it up in Settings → Phone & chats, which lists every channel you have connected and lets you pick the chat and whether it may listen.

1. Connect WhatsApp or Telegram under Connections.
2. Send a message to the private chat you want to use — on WhatsApp that is the Message yourself chat.
3. Pick that chat in Settings. Taskuary names it as the Assistant chat and switches on the assistant for it.

The Assistant then offers Walk me through them in WhatsApp (or Telegram) for each chat set up this way — on its own page and on Phone & chats. Which chat the Assistant offers in the same section picks one — WhatsApp, Telegram, or both.

Only messages sent by the linked owner account, in that exact chat, are accepted. Taskuary marks its own bridge output so its notifications can never loop back in as if you had typed them.

What it can do#

Ask it in plain words. These are real examples, not a command list:

• "what needs me?"
• "walk through important email"
• "what is outstanding?"
• "what did the coding agents finish?"
• "reply to Ruth that the numbers are coming tomorrow"

Walk through is the one worth knowing. It is the same turn-by-turn walk the Assistant page runs: one unresolved item at a time, one question, and it waits for your answer before moving on. The phone is not given a cut-down version of it — the walk carries out the page's job, including the actions at the end of each item.

Answers you give on the phone land in the desktop conversation too, because it is one conversation with two doors.

Choosing#

The phone is the desktop Assistant in a chat, word for word. Every button the desktop card draws is a numbered line here, and on WhatsApp the same choices come again as a poll under the message, so one tap picks - even when the only choice is Next. A number or a tap runs exactly what the desktop's button runs, and no model reads it. Only words you type go to the model.

Typed words are answered by the Assistant's brain - the same one, and the same model, as the Assistant in the app (Settings → How mail is sorted → Assistant). A Claude Assistant runs at low effort unless its model names one (claude-sonnet-5@medium): a turn is a short answer, and Sonnet's own default is high.

Approving things#

A drafted reply comes to the chat with the draft and the desktop's own choices under it - Send, Send and close, Redraft - as numbers or poll taps. Nothing goes out without that pick: approving on your phone is the same deliberate act as approving on the page. The chat never pings you on its own: it speaks when you write to it or hand it the walk.

What you type goes the same way: anything that sends or closes - a reply, a hand-off, Mark done - asks yes or no first. A small change that sends nothing and closes nothing is done at once and said: a priority, the kind of work, its repository, Remind me with a day, Watch it, or "mark the FYI read".

The morning line#

Off unless you switch it on (Settings → Phone & chats). Once a day, the assistant can then open with a short line about the day ahead — what slipped, what is waiting, today's meetings. It is the morning brief's voice, cut to the length of something you read while walking.

It fires once per day, not once per app start, and it is the same cap the brief itself uses: if today's line has gone out, reopening the app does not produce another.

Handing over#

When a conversation on the phone reaches something better done at a keyboard — a coding session, a long document, a report builder — the assistant hands over rather than pretending. The tab it hands to locks onto that item so you arrive where the conversation left off, and the phone thread says so plainly instead of going quiet.

The reverse is true too: a walk started on the page can be picked up on the phone, because the walk's position belongs to the conversation rather than to the surface you started it on.

Agent channels: a chat into one coding agent#

With Developer mode on, a chat can talk straight to one running coding agent - a WhatsApp or Telegram chat or group, a Slack channel, or a Teams chat. Open a coding task, unfold Talk to this agent from… at the foot of its chat, pick the channel and the chat, and bind it. What you write there goes into the agent as if you had typed it in its box, and its answers come back to the chat.

Every binding is listed in Settings → Phone & chats → Agent channels, with Unbind. Add a channel there binds a chat to the coding agents — all of them, yours alone: say "agents" for the list, "TQ-0123 your words" to tell one, or "land TQ-0123". A repository folder keeps it to the agents working there. Your Assistant chat reaches the coding agents the same way while Developer mode is on, so it needs no binding of its own.

What each channel needs from you: Telegram and Slack cannot say which messages are yours, so the form asks your own numeric Telegram id or your Slack member ID (U…); Teams posts with your own Microsoft sign-in.

You can let teammates in, so a few people can work on one project with the same agent:

• Named people only. You add each teammate by name and by their id on that channel (a phone number on WhatsApp, a member ID on Slack, a work email on Teams). Everyone else in the chat is ignored, and nothing said in a bound chat ever becomes a task in your inbox.
• Their words are theirs. The agent reads "Gail Moreno (teammate)" in front of every line of theirs, with a note that they are not you. The task's chat shows it the same way.
• You hold every approval. Only you can approve a push, land the branch, allow a permission request, or let anything be sent, posted or paid. A teammate typing "yes", "1" or the exact words of your choice does nothing - the chat says only you can answer.
• One agent, one repo. A bound chat talks to that one agent, kept to its project's repo. It has no road to your mail, your other tasks or your files, and Taskuary passes nothing on if the agent's checkout is outside that repo.

When the agent wants permission for something Taskuary's own rules do not settle, it asks you: as buttons on the task's chat, and in the bound chat as a numbered list - reply with the number. "Always" answers are remembered for that task and that kind of step, but never for a push or a send. With no answer in ten minutes the request is refused, and both places say so.

A bound chat is the one place Taskuary speaks without being asked first, because you opened it for that: it carries the agent's answers, what it is waiting on you for, its permission requests and its landing question - nothing about your other work. Unbind takes it away again.

What it will not do#

• It will not speak in a group chat, or in a chat you did not choose. (An agent channel, above, speaks only for its agent.)
• It will not send anything a human has not approved, on any channel.
• It will not change a setting behind your back. A change you ask for comes back as a question for your yes.
• It will not answer someone else in your chat — only the linked owner account is heard.

Sections, For later and browsing, on the phone#

The morning line and the walk offer the rail's sections as choices - Walk Tasks, Walk For later, Walk Reports, Walk FYI - and you stay inside the one you picked until it is empty, whether you press Next or act on the item (Make a task, Send to agent, Remind me, Watch it). When the section runs out it says so, and the walk carries on. Rows under For later say when they come back ("back in 3h"). Tasks under Watching - the ones you keep an eye on for somebody else - are still walked, after Tasks, and Walk Watching opens them on their own; press any move on one to stop watching it. "Watch this" typed is the same Watch it as the button, on the phone and on the desktop.

Connections, Reports, Hub and Settings are choices too - the desktop sidebar's four, as picks:

Each step has Back as a choice. A list longer than a poll can hold (twelve) comes ten at a time with More (N left) and From the top; a number you type still picks anything you can see. Changing anything is words - "run it now", "set it up", "make it hourly" - and comes back as a card for your yes, as in the app.

A report's rows arrive as rows (Lakeview · 112), laid out like any other table on the phone. Telegram gets the same cards, with its own buttons instead of a poll and no bold.