Work

Ask and it looks. "What's my latest email?", "anything from Jack this week?", "what did I send Todd about the invoice?" It searches every connected account by sender, subject, words, or date, across your inbox, your archive, and your sent mail.

Ask it to reply and it drafts. "Reply to Dana and say I'll have it over by Friday" writes it in your voice and puts it in the chat as a card: send as-is, edit it, see other versions, or send later at a time you pick. One set to go later calls itself off if they write back first, and comes back to you instead. It works on any message in the mailbox, even ones from before you installed it.

Or start one from nothing. "Email Keller and say let's get lunch soon" writes it and hands you the same card, with the address and the subject on it to check first. Give it an address and it uses that one. Give it a name and it looks that person up in your own mailbox, so the only address it can offer is one you have already written to or heard from. Finding nobody, it asks you for the address rather than guessing.

Meanwhile it watches. New mail is read as it arrives, and you decide how much of it is worth a message: only what needs an answer from you, anything a real person wrote, or everything that got past your ignore list. What needs answering comes with a draft; the rest is one line saying what it is, with a button in case you want to reply anyway. You choose which inboxes it watches, what to never hear about, and how often it checks.

If you checked Let it tidy up, it can move mail out of your inbox, which is your mail app's archive and puts back in one word. It can never delete anything.

Works with any email account connected on the agent's Connections tab, under Email, calendar, files.

Works with

• an email account

Made by Otian AI.

Nobody drops a promise on purpose. It happens because the promise was made in a sentence, in the middle of something else, and never written anywhere. This reads your sent mail and finds the commitments you made: what you said you would do, who is expecting it, and by when. Each one becomes a tracked item you can see in one list, and it tells you before the deadline rather than reporting the miss afterwards. With Meeting Notes installed, the daily sweep also reads your transcripts, so the promises you made out loud get caught too. Waiting On tracks what other people owe you; this is the other half, the one that protects your name.

Works with any email account connected on the agent's Connections tab, under Email, calendar, files.

Works with

• an email account

Made by Otian AI.

This is the run where the catching actually happens. Each morning it re-reads your recent sent mail, and because it loads Meeting Notes alongside when you have it, this is also the only time your recorded meetings get searched for promises. Anything new gets tracked, and anything due tomorrow or today gets said out loud. A promise that has gone late is raised the day after, then at three days, a week, and two weeks, rather than every single morning: a reminder repeated daily teaches you to ignore the messenger, and the item stays in your list whenever you ask. A morning with nothing owed stays silent.

Works with

• Commitment Keeper skill

Made by Otian AI.

You send the email, then carry the invisible job of remembering that no answer ever came. Same with the invoice. This hands both over: your agent scans your sent mail for real asks and real invoices, checks whether each one got a reply or a payment, and keeps one live list of what you're still owed. Ask who owes you and get the money first, longest overdue at the top, with the outstanding total and a chase drafted in the right register: light for four days late, plain and specific at a month. Tell it “that one's fine” and it drops the thread for good. It reads your mail and drafts your words; sending stays yours, and it never takes a payment or names a consequence you didn't ask it to.

Works with any email account connected on the agent's Connections tab, under Email, calendar, files.

Works with

• an email account

Made by Otian AI.

The Waiting On list only helps if someone keeps it fresh. This does: each morning it re-reads your recent sent mail, marks the asks that got answered, spots the ones that crossed the too-quiet line, and stays silent when nothing moved. No news really is no news.

Works with

• Waiting On skill

Made by Otian AI.

One to-do home is enough: if your real list lives in Todoist or Google Tasks, install that skill instead of this one, or "add a task" has two places to land. Turn your agent into a to-do list that actually sticks. Say "remind me to send the invoice Friday" and it files the task with a due date and sets the reminder for Friday, so the day brings a message and not just a row you have to go and look at. Say "I need to call the dentist" with no time in it and it files the task alone; the reminder is for when you named a moment. Ask "what's on my plate?" and it posts a numbered checklist with a check button beside each item, so finishing something is one tap rather than a sentence. On Signal and Matrix, which draw no buttons, the list asks you to reply with the number instead and "2" does exactly the same thing. Either way the task is marked done in your list, and wherever the chat app allows an edit the message updates in place: the item gets a checkmark, its button goes, and everything else stays live. Every task is saved to a real list, so nothing is forgotten the moment the conversation scrolls away.

Made by Otian AI.

Hand it the writing. It drafts emails, social posts, reports, and summaries, or takes a second pass at something you already started, in your agent's own voice. It never sends or posts anything: every draft comes back to you first.

Made by Otian AI.

Hand off anything you'd otherwise go and look up. The Researcher searches the live web for current prices, recent news, product documentation, and who works where now, then reports back with the source URL for every claim. Without it, your agent answers from training data alone and cannot tell you when that data is out of date.

Made by Otian AI.

Every day at your set time, the agent posts your open tasks: what's overdue first, then what's due today, then what's coming. Each item has its own check button, so finishing one is a single tap; on Signal and Matrix, which draw no buttons, you reply with the number instead and it counts the same. Wherever the chat app allows an edit, the message updates in place as you go, and everything you leave is still there tomorrow. No app to open; it just arrives in your chat.

Works with

• Task Manager skill

Made by Otian AI.

Every Sunday the agent reviews your Tasks and sends a short recap of what got done, what's still open or overdue, and what's coming up next week, so nothing carries over unnoticed.

Works with

• Task Manager skill

Made by Otian AI.

Your agent reads your calendar, on Google or Outlook, and drafts changes to it in chat. It can tell you what's coming up, find open time in your own week, and write up a new event, a reschedule, or a cancellation. Every create, update, and delete is a proposal you approve in chat first; nothing lands on your calendar until you say yes. Two limits worth knowing before you install. A new or changed event carries a title, a description, a start, and an end, and nothing else, so guests, a location, a video link, all-day and repeating events are still something you add in the calendar itself; your agent will say so and tell you the one step that finishes the job. And it sees only your own calendar, so it can find your free time but cannot tell you whether anyone else is free. Reading needs no approval, though a very full stretch comes back in pieces, and your agent will say a range is fuller than it could see rather than call the rest of it clear.

Works with

• a calendar

Made by Otian AI.

Open the page for Calendar Manager →

Each morning it reads today's calendar and sends one line per event: time, title, and where. It adds a prep note only when the event's own notes say something to do. An empty day is one line.

Works with

• Calendar Management skill

Made by Otian AI.

Open the page for Today's Calendar →

One to-do home is enough: pick this, Task Manager, or Todoist, whichever holds your real list, or "add a task" has two places to land. Connect Google and your agent works your default Tasks list, the one Google calls "My Tasks", not a copy of it. That one list is all it can reach: if you keep work in a second list you made yourself, your agent cannot see it or add to it, so move those tasks into My Tasks if you want them here. Say "remind me to send the invoice Friday" and it lands there with a due date and the reminder is set for Friday; ask "what's on my plate?" and it reads back what's actually on the list; say "done with the invoice" and it checks it off right away. Changing a task's name or date happens right away too, and you can undo either one: open the job on the Jobs tab and press Undo, or tell your agent to put it back. Deleting a task is the one change it asks you about first, because a deleted task is gone for good. Nothing is mirrored into Archie, so there is no copy to drift out of date. Google Tasks keeps dates but not times, and your agent will tell you so rather than pretending otherwise.

Works with

• Google Tasks

Made by Otian AI.

The Researcher answers a question. This goes and does the work. Give it something you would otherwise spend an evening on, such as which of these three suppliers to use, what the rules actually are where you live, or whether a claim you read holds up, and it searches from several angles, opens the pages rather than trusting the search snippet, weighs what disagrees, and comes back with a written answer and a source for every claim.

It is deliberately slower and costs more than the Researcher: up to 20 searches on your own AI account, each about 2 to 4 cents with the pages it brings back, which comes to about 50 cents a run. Use the Researcher for a quick fact and this for a decision you are actually going to make.

Made by Otian AI.

Scheduling with another person is a small negotiation: offer slots, wait, read the counter, check the calendar again, finally book it, and remember all of it in between. This runs that whole loop. It picks real gaps from your calendar (respecting standing rules like "no meetings before 10", said once), writes the offer in your voice for you to send, and notices their reply when it arrives rather than waiting to be asked, so a slot agreed at ten in the morning is staged for your approval at ten in the morning. Several negotiations can run at once and none of them lives in your head. Finding your own free hour is the Calendar Manager's job; this skill starts the moment someone else's yes is required.

Works with any email account connected on the agent's Connections tab, under Email, calendar, files.

Works with

• an email account
• a calendar

Made by Otian AI.

The whole point of handing scheduling over is not having to check for the answer yourself. This does the checking: every morning it looks for replies on each open negotiation, tells you a time was agreed so you can reply "book it", which stages the event for your approval, relays a counter-offer with a checked response ready to send, and stays completely silent when nothing moved.

Works with

• Find A Time skill

Made by Otian AI.

Somebody has to open the day: what's on, who's got what, what's still owed. Usually that somebody is the owner, every morning, from memory. This routine does the assembling instead. Each morning the agent posts one message, built from what it can actually read: the shared calendar's day, and, when those add-ons are installed, the handoff ledger's open and unclaimed work and the customers still waiting on a reply, oldest first. Sections with nothing to say are left out silently, so a quiet day is a short message rather than a wall of “no updates”. It reads the accounts the business shares and never a teammate's own mail or calendar, which stays between that teammate and the agent. The time is editable, and the brief is a report, not a ritual: nobody is asked to answer it.

It lands in the one chat your agent posts in, named at the top of the Routines tab. That chat is always between you and the agent, never a group, so passing the brief to the team is yours to do.

Works with

• Calendar Management skill

Made by Otian AI.

One to-do home is enough: pick this, Task Manager, or Google Tasks, whichever holds your real list, or "add a task" has two places to land. Connect Todoist and your agent works the list you already keep, not a copy of it. Say "remind me to send the invoice Friday" and it lands in Todoist with a due date and the reminder is set for Friday, so the day brings a message and not just a row; ask "what's on my plate today?" and it reads back what's actually there; say "done with the invoice" and it checks it off right away. Changing a task's name or due date happens right away too, and you can undo either one: open the job on the Jobs tab and press Undo, or tell your agent to put it back. Deleting a task is the one change it asks you about first, because a deleted task is gone for good. Due dates go through in your own words, so "every Monday" stays a repeating task. Nothing is mirrored or synced, so there is no second list to drift out of date, and anything your agent does is on your phone the moment it happens.

Works with

• Todoist

Made by Otian AI.

Meeting Prep can always tell you who you are about to meet and what it is about. What it cannot do on its own is arrive, so you have to remember to ask, which is the thing a briefing is for. This one watches the calendar and sends the prep thirty minutes before each meeting starts, so it lands while there is still time to read it. Change the notice to whatever you actually need. An all-day entry never triggers it, because a birthday is not a meeting, and neither does anything already canceled.

Works with

• Meeting Prep skill

Made by Otian AI.

Before your next call, ask for a prep. Your agent reads the event straight off your calendar, with who's invited, when and where, and anything in the invite, and turns it into a quick brief with two or three sharp things worth raising. Ask "what's my day look like" for the whole day at a glance, back-to-backs flagged. It only reads your calendar; anything you want changed goes through the Calendar Manager. With Email Manager (or another skill that reads your mail) installed, the brief also carries your last exchange with the people in the room: what was promised, what's still owed.

Works with any calendar account connected on the agent's Connections tab, under Email, calendar, files.

Works with

• a calendar

Made by Otian AI.

By Monday it is too late to fix an overcommitted week; by Friday it is still just an email. This does the arithmetic ahead of the weekend: meetings and travel time against your working hours, plus the work due that nobody has booked time for. Then it says plainly whether next week fits, and when it does not, names the specific things worth moving and why. A week that genuinely fits gets one line telling you so and where the slack is.

Works with

• Honest Week skill

Made by Otian AI.

Connect Fireflies, Zoom or Calendly and your agent quietly builds a memory of every meeting you have. It pulls new transcripts, works out which speaker is which person, writes a summary with the decisions and action items, and files it away. Then you can just ask: what have Sarah and I talked about recently? What did we decide about pricing? What did I commit to on Tuesday's call? When a transcript comes back with anonymous 'Speaker 1' labels, which Fireflies does for uploads, Teams, and phone calls, the agent infers who's who from the transcript and the invite, shows you its guess, and remembers your correction for good.

Works with

• Fireflies
• Zoom
• Calendly

Made by Otian AI.

A call ends, the recorder writes the transcript, and this files it and hands you the short version: who was on it, what it was about, the decisions, and the action items with owners. It asks the recorder every fifteen minutes whether anything new is there, which costs nothing, and only does real work when something is. If it had to guess who was speaking, it shows you the mapping it inferred and asks you to confirm it before that becomes the record.

Works with

• Fireflies skill

Made by Otian AI.

Everyone plans a week as though it has infinite hours, then spends Friday wondering what happened. This reads your calendar and your task list together, counts the hours honestly, including the ones eaten by back-to-backs and travel between things, and says the quiet part out loud: you are overcommitted by six hours, and here is what to cut. It names specific candidates and why (a recurring meeting you have skipped three times running, a call that could be a message), then stages the moves for your approval. It also finds the hours you have and nobody has claimed, and offers to defend them before someone else does. It never changes your calendar on its own.

Works with any calendar account connected on the agent's Connections tab, under Email, calendar, files.

Works with

• a calendar

Made by Otian AI.

Small teams lose work in the gap between “someone should” and “I did”, and the owner ends up as the router who remembers who took what. This hands that over. When somebody claims a piece of work in the team's chat, the agent files it: what it is, who has it, when it was raised, and the due date if one was said. When work is raised and nobody takes it, that gets a row too, with the owner blank, and blank owners lead every report because unclaimed is where work goes to die. Ask “what's open?”, “what's on Dana?”, or “who's got the Meyer quote?” and get one line per row, longest-open first, with ages counted honestly. Handing a thing to someone else moves the row and keeps the history. Done is said, not assumed: a row closes when its owner says so, and until then its age is the fact you see. The agent never nags anyone; it keeps the ledger true and reports it when asked.

Made by Otian AI.

An Airtable base starts as a good idea and turns into a museum. Rows go in and nobody closes them, a status column stops matching what actually happened, and the view you built in March answers a question you no longer ask.

The work is not typing rows. It is keeping what you already typed true, and that is the part nobody does.

Connect Airtable and your agent works the real base. Ask what is in a table and it reads the live rows. Tell it to file something and it finds the table, reads the columns that actually exist, and builds the row from those instead of guessing at names. Ask what has gone stale and it sorts by when each row last changed and tells you what nobody has touched.

It remembers where things live, so the second question about a table costs no searching. Changes are shown to you before they happen.

Works with

• Airtable

Setting up Airtable Keeper

1. Make the token at Airtable, at airtable.com/create/tokens. Check schema.bases:read and data.records:read, plus data.records:write if you want your agent to be able to add or change a row.
2. On the same screen, under Access, add each base you want your agent to work with. A token with every scope and no bases added can see nothing at all, and every call comes back refused. This is the step people miss.

Made by Otian AI.

The list only stays true if somebody keeps it true, and during a hunt that somebody is the person with the least attention to spare. This runs once a week, reads the confirmations, scheduling mail, rejections and offers that arrived since the last run, and moves every row to the stage it actually reached.

Then it tells you the part you need: anything waiting on an answer from you, what moved, and what has gone quiet in a way that means something. A thread that went silent after an interview is a real signal; forty applications nobody replied to is a Tuesday, and it says how many there are rather than listing them.

Weekly rather than daily, because the mail already reaches you the moment it arrives. What this does is keep the list honest and tell you the shape, and neither of those changes enough in a day to be worth a message. A week with nothing in it is one line.

It drafts follow-ups and never sends one, and it will not apply for anything or answer an offer.

Works with

• Job Hunt skill

Made by Otian AI.

Forty fields, and you have answered every one of them before. Address, dates, policy numbers, the same three paragraphs about what you do. Forms are a pure tax, and the reason nobody automates them is that every form is different.

This fills them the way a person would: it reads the actual page, works out what each field is asking for, and answers from what your agent already knows about you. Where it does not know something, it asks you once and remembers, so the next form that asks is free.

It always stops at Submit. You see the finished screen before anything becomes real, and pressing the button is yours. It never touches password or payment fields, and if a form wants a sign-in code it hands the window straight back to you.

That is also where it compounds: a form is a questionnaire about your life, so every one it fills teaches it facts it did not have, and it gets faster at the next one on its own.

What Form Filler needs

Needs Computer control turned on, which is off until you switch it on. Your agent works the site in a browser window on your own computer, and you can watch it. It never types a password or a card you pay with, and it asks before pressing anything that finalizes.

Made by Otian AI.

The work of having repositories is not writing code, it is the checking: what did someone say on that pull request, which review has been sitting for a week, which issue did you say you would look at and then not. GitHub notifies you about all of it and the notifications become weather.

Connect GitHub and your agent reads what is actually there. Ask "what's waiting on me?" and it lists the reviews requested of you, the issues assigned to you, and the pull requests where someone replied and you never answered, oldest first. Ask "what's gone quiet?" and it finds the pull requests with no activity for a week that nobody has closed. Ask what a change does and it reads the diff and tells you in plain sentences.

It can write too, and writing is always shown to you first: a comment, a label, closing something. Nothing reaches a repository other people can see without you saying yes.

Works with

• Github

Made by Otian AI.

The work of having projects is not writing code, it is the checking: which review has been sitting since Tuesday, what did someone say on that merge request, which issue did you say you would take and then not. GitLab emails you about all of it and the emails become weather.

Connect GitLab and your agent reads what is actually there. Ask what is waiting on you and it lists the merge requests where you are the reviewer, the issues assigned to you, and your own merge requests where somebody replied and you never answered, oldest first. Ask what has gone quiet and it finds the open merge requests nobody has touched in a week. Ask who still has to review something and it diffs the reviewers against who has actually approved.

It can comment too, and a comment is always shown to you first. Nothing lands on a project other people can see without you saying yes.

Works with

• Gitlab

Setting up GitLab Keeper

1. Make the token at GitLab, under your avatar, Preferences, Access tokens. Check read_api if you only want your agent to read, or api if you want it to be able to post a comment.
2. The read-only scope cannot comment, and when it tries, the failure reads like a permission problem on the project rather than on the token. Make a new token with api checked and connect that one.

Made by Otian AI.

It opens by counting, not by asking. The first thing it does is go through your inbox and come back with the piles: how many from each sender, how many you never opened. Nobody knows they have 214 emails from one bank, which is why the tidying never gets done.

Then it asks about the big ones, one at a time. File these away, leave them alone, or tell you when they arrive. Whatever you answer becomes a standing rule you can see and change in the Your inbox rules list on the Knowledge tab, and it never asks about that sender again.

After that it keeps watch. Each morning it applies the rules you set, tells you what it moved, and asks about senders it has never seen before. Filing away is the archive your mail app already has: the message leaves the inbox, stays where it can still be searched, and goes straight back, so a wrong call costs you one sentence. It cannot delete anything and it cannot send anything.

Works with any email account connected on the agent's Connections tab, under Email, calendar, files.

Works with

• an email account

Made by Otian AI.

Deciding what should happen to a pile of mail is worth doing once. Doing it again every week is the thing you installed an agent to stop doing.

This is the second half of Inbox Rules: it takes the decisions you already made and keeps them running. Each morning it files away whatever arrived from the senders you said to file, and reports the count in one line, with the offer to put any of it back. Then it names anyone new, one sender at a time, and asks what you want done. Whatever you answer becomes another standing rule.

A morning with nothing to move and nobody new stays silent.

Works with

• Inbox Rules skill

Made by Otian AI.

A hunt is thirty applications across eight hiring systems, each with its own login, and a mailbox where the confirmation, the interview invite and the rejection for the same job are four weeks apart. Nobody can see the shape of that from inside it. This keeps one list: what you applied for, where each one got to, and what the posting actually said, saved before the listing comes down and takes the requirements and the salary band with it.

It fills the list from your mail rather than asking you to. Every hiring system sends a confirmation when you apply, a scheduling mail when they want to talk, and a rejection when they do not, and it reads all three and moves the row itself. Ask how it is going and you get the count at each stage and, if the numbers are big enough to mean anything, which stage is the one leaking: many applications and no screens is a different problem from screens that never become offers, and only one of them is fixed by applying to more places.

It drafts follow-ups and never sends them, and it will not apply for anything or answer an offer on your behalf.

Reading your mail needs an email account connected on the agent's Connections tab, under Email, calendar, files. Without one it still keeps the list from what you tell it.

Works with

• an email account

Made by Otian AI.

Linear is where the work is written down, and the cost of that is a tab you have to go and open. What is assigned to me, what moved while I was in meetings, what is due Friday, what did somebody say on that issue: every one of those is a trip to a board and a bit of scanning, and the answer is usually four lines long.

Connect Linear and your agent reads the board instead. Ask what is waiting on you and it comes back with the issues assigned to you that are not finished, in the order they last moved, with the status in your team's own words and the ones that are late at the top. Ask what changed this week and it says what changed and who changed it. Ask about one issue and it reads the description and the comments and tells you where it stands.

It can open an issue, move one, reassign one, and leave a comment too, and every one of those is shown to you before it happens. Nothing lands on your team's board without you saying yes.

Works with

• Linear

Setting up Linear Keeper

1. Make the key in Linear, under Settings, Account, Security and access, then Personal API keys. Give it Read plus Write if you want your agent to be able to propose a change; Read alone answers every question here.
2. A key can be limited to certain teams. One that is will answer about those teams and nothing else, and an issue outside them comes back as a permission error rather than as missing.
3. Paste it in Archie on the Connections tab, then Apps, beside Linear.

Made by Otian AI.

Every new person costs the owner the same hour: here's the agent, here's what we use it for, here's where things live, ask me if you're stuck. This hands that over to the agent itself, which is who a new teammate meets first anyway. When someone new appears in the chat, the agent introduces itself, learns their name and role, and gives a tour that is three beats long: what this team actually asks it for, what connecting their own mail and calendar gets them (answers from their own accounts, readable by nobody else on the team through the agent), and where the shared lists live. Questions about how the business does things are answered from the files under Knowledge, with the file named, and when nothing covers a question the agent says so instead of improvising policy. The welcome ends with something real, a calendar check or a first filed handoff, because a tour that ends with a tour is a brochure. It runs once per person; returning teammates get a colleague, not a greeter.

Made by Otian AI.

Notion is where good intentions go to sit. Pages get made and never finished, statuses stay on "In progress" for four months, and the database you set up in January is now a museum. The work is not writing things down; it is keeping what you wrote down true.

Connect Notion and your agent works the workspace you already have, not a copy of it. Say "put this in my reading list" and it finds the right database, reads what columns it actually has, and fills them in. Ask "what's in my project tracker?" and it reads back what is really there. Say "mark the Henderson project done" and it shows you the exact change before it touches anything. Ask "what's gone quiet in Notion?" and it tells you which rows nobody has touched in weeks, which is the question no Notion view answers well.

The first time you use it, it looks around and shows you what it can see, so you are never guessing whether it is connected. Nothing is mirrored into Archie, so there is no second copy to drift out of date.

Works with

• Notion

Setting up Notion Keeper

1. After connecting, open each Notion page or database you want your agent to work with, click the three-dot menu at the top right, choose Connections, and add your integration. A Notion key on its own grants nothing: your agent sees only what you share here, and this is the step people miss.

Made by Otian AI.

Exec Concierge is Concise with a point of view. It leads with the answer, keeps every reply scannable, and then does the thing a plain summary will not: when there is a choice on the table it says which option it would take and gives the one line of reasoning behind it, because you asked precisely so you would not have to weigh it yourself. It filters as hard as it reports. Whatever it can handle without you gets handled and mentioned in a clause, and only the items that genuinely need a decision get a line of their own. Ask for a draft and the draft is in the reply, not a promise of one. Anything that cannot be taken back stops first: sending, deleting, declining and spending each get one short line and a wait.

Exec Concierge in a conversation

Made by Otian AI.