What's new on OTP

• What OpenAI shipped. Dots are agents with their own computer and browser that keep working after you close the tab, reaching thousands of apps through ChatGPT plugins. Space and Pages give a team shared files and documents. Team Tasks puts recurring work on a schedule. A Meetings plugin turns a call into notes and action items. Dots are on the Pro and Business Premium plans for now, outside Europe and the UK.
• What it leaves out. A dot can do a task. Nothing in ChatGPT says which seat it holds, which number it owns, who it answers to, or whether last quarter's decision worked. That is the organization, and the organization is what OTP records.
• Meeting notes are becoming free. Good. Wherever your notes come from, ChatGPT, Fellow or Fireflies, the value is in what happens next: the to-do on an owner, the number on The Board, the decision checked against the result. You can already bring a transcript into a meeting on the Meetings page (AI meeting follow-ups, in Labs) and Ollie drafts the follow-ups.
• Where OTP is headed. OTP already speaks MCP, the protocol ChatGPT plugins are built on, and agents already hold seats on the chart and post Signals to the meetings they report to. Next: OTP inside ChatGPT, so an agent can read its own seat, numbers and to-dos before it starts work. After that, OTP raising an event the moment a to-do lands on an agent's seat, so the agent starts without being asked.
• A ceiling before anything runs all night. An agent that works around the clock can spend around the clock. Before any always-on agent can reach your organization, each connection gets its own spending cap, on top of your wallet balance. Your AI, your ceiling.

• Type what you want to do. "Add meeting", "invite", "top up". Each result shows the path you would click, Meetings > + New meeting, so it teaches the app while it takes you there. Choosing a result opens the page; nothing is created or changed for you.
• Search your own work. Meeting titles, summaries and notes, to-dos, issues, quarterly priorities, projects, headlines and your notepad.
• Only what you could already see. Results follow the same rules as the lists they come from: your own to-dos, your teams' issues and meetings, a private issue or priority only if it is yours, and other people's notes only when they shared them and Shared notes is on.
• Not yet. Meeting transcripts are not searched; their summaries are.

• Any notetaker. Export from Zoom, Teams, Fireflies, Fellow, ChatGPT or anything else that writes a file, and upload it.
• Named for you. The meeting title defaults to the file name.
• A clear limit. A transcript over 500,000 characters now says so plainly instead of failing with a generic error.
• In Labs. Part of AI meeting follow-ups. Turn it on under Settings, Labs.

• Signed in as. Your email is on every setup screen, so the wrong account is obvious before you build anything with it.
• Use a different account. Signs you out first, so Google does not hand back the same account.
• Private by design. The warning never shows the other email address or the company name.

• Agent activity this month. Calls made, included calls used, billable calls and their cost, plus what the calls actually did.
• The price. 500 calls per organization per month are included on every plan. After that it is $10 per 1,000 calls, taken from your wallet once a day. One request from your AI counts once, even when OTP does several things to answer it.
• Balance emails. One email each time you cross a line: credit at $5 or less (skipped if auto-recharge is on), a balance below zero, and a balance past -$5, where new AI work pauses until credit is added.
• What never pauses. Meetings, to-dos, priorities and The Board are free and keep working whatever the balance. Only AI work waits: Ollie, the meeting bot, recordings and transcripts, and agent processes.

• Edit meeting to-dos. Title, description, due date, priority and owner, from the same form as your personal to-dos. It stays a meeting to-do after you save.
• Hand it on by name. Tell your AI "give this one to Owen" and it does. Over MCP, add_todo now takes a person's name and a new update_todo can edit or reassign. A full name always works; a first name works when only one person has it. OTP never guesses: an unknown or ambiguous name writes nothing and says why.
• Cleaning up the old ones. A to-do whose title starts with a name in brackets, like "[Owen Cartier] send the deck", now shows the title clean with one click to assign it to that person.
• Every to-do on a person and a team. The create form has a Team field. Pick a meeting and it fills in that meeting's team; pick a team without a meeting and it is a team to-do; leave both empty and it is personal.
• Start timer, not Start. The button beside Done tracks time. It now says so.

• Meeting scores read your numbers. The score treated every KPI as blank and every team as having no priorities set, so it came out lower than it should have. It now reads KPI values and priorities, including which ones are off track.
• Before the meeting stays current. The brief is prepared the day before. The meeting page now hides anything resolved since: a priority completed, a to-do done or deleted, a KPI retired or moved to another team.
• Due dates stay put. Editing a to-do could shift its due date by your time zone offset on every save. It now keeps the date you chose.

• Written on its own. Ending a meeting, by hand or when an abandoned meeting closes itself, now produces the insight if the meeting has something to read. If the transcript arrives late, Ollie tries again when the next meeting in the series starts, so the "From your last meeting" card is filled when you open it.
• At most once per meeting, and it costs what the button cost. It is the same Ollie call as pressing Generate. It counts toward your first five free meetings, then draws on your wallet. It never runs with AI switched off or an empty wallet.
• Listen. A play button on the insight, in the meeting and on the follow-ups page. It uses the voice built into your browser: nothing is charged and no text leaves your machine. The voice sounds different on a Mac, a Windows laptop and a phone, and the button stays hidden on a browser that cannot speak.
• In Labs, on by default. "Ollie's insight, without asking" and "Hear Ollie's insight" are both switched on for every organization and can be turned off in Labs.

• Unticked by default. To-dos, issues, quarterly priorities and numbers Ollie suggests all start unselected. Anything you add by hand still starts ticked, because typing it in was the choice.
• Grouped, with counts. Each kind has a header showing how many Ollie suggested, with its own Select all and Select none.
• Create selected (N). The button counts your picks as you tick them. Rows are more compact, and the ones you pick turn bold so they stand out.

• Held, not dropped. Before the meeting is in progress, updates from others wait behind a chip in the corner that says how many came in, with a Show button. Everything is already saved; only the redraw waits. The moment the meeting starts, everyone lands on the same state, and a running meeting still moves together.
• Before this meeting is a step. The prep brief, what your own records say is worth the time, now sits in the agenda as the first step, ahead of Warm-up, instead of pinned above every section.
• Focus mode means focus. It now hides everything except the active step. Before, "Carried from last session" in a 1-on-1, and a second copy of a section a custom agenda used twice, stayed on screen for every step.

• Private unless you share. Every note, including every note written before this existed, stays private. Sharing is its own deliberate click, never a side effect of saving.
• Read-only for everyone else. A shared note opens with a line naming who wrote it. Only the author can edit, unshare or delete it.
• Whole organization only, for now. There is no per-person or per-team sharing yet, so do not put anything in a shared note you would not say in an all-hands.

• Invite a team member from the avatar menu. It used to offer Add an agent and nothing for people. Roles that can invite now see the link.
• Ongoing projects. A project's due date has a No due date button, and the form says to leave it blank for an ongoing project.
• Name first, then create. New project used to create an untitled project on every click. Nothing is written now until you type a name.
• Teams and Projects are not the same thing. Choosing the "project" team type now points you to Projects, where project work lives.
• One number, one place. Creating a second number with the same name for the same owner is refused, with a pointer to the one that exists. Duplicates meant someone entered a score into a copy nobody's board was reading.

• What Ollie sees. Meeting search, summaries, transcripts, participants, and action items, through Fellow's own MCP server. Ask Ollie what was decided on Tuesday and it reads the note.
• Read-only by construction. The connection asks Fellow for read scopes only. Agenda edits, template changes, and deletes are never requested, and OTP filters those tools out of Ollie's hands as well.
• Your Fellow permissions apply. Fellow only returns meetings the connecting account can already see. A workspace admin must allow MCP connections in Fellow's security settings first.
• Also fixed on the way. Some connected tools, GoHighLevel among them, were resolving to an empty tool list. Ollie now asks Composio for each toolkit's current version, so those tools mount again.

• What you run. Pick your operating system from the template library, or say none yet and answer three questions: how many people, how often, and what the meeting is mostly for.
• The proposal. The weekly meeting that system prescribes, or OTP's own Delta Meeting, with alternatives one click away. The people, read from who you actually meet with when your calendar is connected, otherwise from your members. The slot your team already keeps, otherwise Monday at nine.
• Approve once. Schedule it records your system on the organization, saves the agenda as one of your formats, adds the ticked people to the team, and creates the series with the agenda and attendees filled in. Nothing is created before that.
• Where it appears. The onboarding finale, the Ollie Home first-meeting card, the first-meeting Flow, and a new card on the Meetings page when your weekly meeting is not on the calendar yet.
• In Labs. Rides on The new OTP flag. Off, every one of those links goes to the Meetings page as before.

• One page, one search. Your own formats sit on top, the 181 public templates from orgtp.com/templates sit under them, and the search box in the hero looks through both.
• Run now. Every template starts a meeting in one click, with the agenda snapshotted onto that meeting. Steps that name your numbers, priorities, issues or to-dos become the live sections, so your real numbers show up in the room.
• Customize. Open any template in the builder, change the sections and timeboxes, and save it as one of your formats. The builder, Make default, Edit and Delete are unchanged.
• In the Start a meeting modal. The Agenda picker gains Choose from the template library: type a word, pick the agenda, and schedule it with the same date, repeat, team and video-link fields as any meeting. Schedule on any template card opens that modal with the agenda already chosen.
• From the public page too. Signed in, every template on orgtp.com/templates has Run this template now and Schedule it.
• In Labs. Rides on The new OTP flag. Off, the Meeting formats page is exactly as it was.

• Tick and send. Connect a directory once and the dialog shows everyone in it, with the people already in OTP greyed out. Tick the rest and one click sends every invite.
• A link you can paste anywhere. Copy the invite link into Slack or an onboarding doc. Anyone who opens it, signs in, and has an email on your organization's domain joins as a team member. Regenerate it and the old one stops working.
• Auto-add. Switch it on and anyone who signs in with an email on your domain is added without waiting for an invite. Off by default.
• On for everyone. The full invite form with roles, seats and access toggles is still there under Advanced invite for the cases that need it.

• In the main menu. Whiteboard and Friday Reconciliation now sit in the main menu instead of the Labs section.
• Before the meeting. Every organization's meetings get a prep brief, and attendees get it by email the day before.
• Ask Ollie in a meeting. @ollie answers in the meeting's bot chat for every organization, charged to your wallet like any Ollie answer.
• Nothing else changed. Graduating means no switch, not free: anything metered is metered exactly as before.

• Answers, not prompts. Five tiles say what is true right now, such as how many numbers are outside goal, which priorities are off track and how many to-dos are past due. The face of a tile is plain arithmetic over your records, so reading Home costs nothing. A tile that cannot show its evidence does not render, and "all on goal" is an answer, not a blank. People who run the company see the company; everyone else sees their own work.
• Ollie does the work, you approve it. On Priorities, Ollie drafts one concrete next step for each priority that has none, and you edit, tick and save the ones you want. On Last week, it drafts follow-ups meeting by meeting. On Follow-through, it raises the past-due pile for the next meeting as one issue, without chasing anybody. Drafting is charged at actual cost from your wallet; nothing is written until you say so.
• Ask on Home itself. The chat lives on the page now, with your saved conversations beside it.
• Keep watching a tile. Put any tile on a weekday or Monday morning check in your own timezone. When nothing has changed it stays silent and costs nothing. When something has, Ollie leaves one short read on Home.
• Flows. A library of sequences for the things teams set up, like a weekly meeting or numbers with owners, each step carrying the reason it comes where it does. An admin can drag the steps into their own order. Flows use your organization's own words, so if you call your priorities something else, the flows use your word.
• The company page: what we said, against what is true. For the people who run the company, it puts your Operating Plan beside your actual numbers, priorities, to-dos and seats, and ranks what does not match, each finding naming what it looked at. Some fix themselves from the page: close out someone who has left by moving their open work to their manager, put the number your plan calls defining onto The Board without inventing a goal for it, or raise the overdue pile as one issue. A company with nothing off is told "Nothing is off."
• Work shows what the work cost. What Ollie did for you, with the total and the split between meetings, answers and everything else.
• Just for you, first. Labs now has a "Just for certain people" list on each beta, and anyone can switch a beta on or off for themselves without waiting for an admin. For a new organization, the menu also starts shorter and grows as you set things up, so nothing you have not reached yet is in the way.

• The recap, Ollie's read and the pre-meeting brief reach everyone in the room. All three decide who to email through one function, and it only recognised one of the three ways a meeting stores its attendees. When it recognised none, it sent to nobody and said nothing. Measured across every meeting with attendees, only 36 of 94 could reach anyone before the fix. All 94 can now. Teammates who have not claimed a seat are matched on their membership, never on a guessed address.
• A red bar when nothing is recording. On a live meeting with no Ollie capture and no recorder running, a slim red bar reads "This meeting is not being recorded." If Ollie was capturing earlier and has left the call, it says that instead, because a bot that drops out mid-meeting is otherwise discovered when the notes never arrive. From the bar you can record in the page, send Ollie back to the call, or mark it as not recording on purpose.
• Follow-ups without a transcript. A meeting run live in OTP with no audio used to be refused. Ollie now reads what the meeting actually holds: section notes, Signals, issues solved in it, to-dos created in it, the numbers and priorities as they stood, and the ratings. It is told plainly that there is no transcript, so it does not invent dialogue or quotes.
• "Ollie was not sent to this meeting." When auto-join skipped a meeting, the meeting sat at scheduled forever. It now carries an amber row saying so, with a link to your meeting presence settings.

• Done or not. "When this one is done" waits for you, and can count the next due date from the original schedule or from the day you finished. "On schedule" keeps coming whether or not the last one was ticked.
• A start and an end. Starts on a date you choose. Ends never, on a date, or after a number of times.
• A preview before you commit. The picker writes the rule back as a sentence and lists the next five dates, computed by the same code that saves it, so what you read is what you get.
• Editing a copy asks what you mean. Just this one, or this and all future ones. Setting a copy to not repeat stops the series and keeps the copy.
• It runs on your clock. Rules used to run in UTC, so a Friday evening to-do on the West Coast repeated on Saturdays. Dates now follow your own timezone, across daylight saving too.
• Shared to-dos behave. One owner ticking their part no longer creates the next copy while the to-do is still open, and co-owners carry over to the next copy instead of dropping off.

• The agenda is not hidden behind "advanced" any more. It used to default to record only, which is how whole recurring series were born without an agenda.
• Today is your today. The block uses your browser's timezone, so at nine in the evening on the East Coast tomorrow's meetings do not show up under Today. An empty day shows nothing rather than promoting tomorrow.
• Calculated numbers are current when the room reads them. A formula edited after its inputs existed, or an import, could leave a stored value behind the formula. Opening a live meeting now recalculates them first. A meeting that has already frozen its numbers keeps them exactly as they were, because that snapshot is what the discussion was about.

• Private by construction. Another person in your organization sees nothing, and neither does the same person looking from a different organization. A conversation you delete stays deleted even if an old browser tries to import it again.
• If your account cannot be reached, the chat says the conversation is saved on this device only, rather than pretending.
• "What did we decide last week?" works for big organizations. Ollie's meeting list was reading every meeting's full transcript at once, which for an organization with a long history was more text than the model can take, so the question could fail every time it was asked. It now reads a short summary of each meeting and opens a transcript only when it needs that one.

• Four tabs: Compliance, Controls, Data and infrastructure, and Questions. Deep links work, so a reviewer can be sent straight to the controls.
• A gap is shown as a gap. Database row-level security and application-level column encryption are marked not in place, each with what covers it today.
• The "last updated" date cannot go stale. It is generated from the history of the files that hold the claims, and a change to a claim that leaves the date behind cannot be merged.
• Nothing third party watches you work. The pages you see once signed in, including a private meeting page whose title is your own content, no longer load analytics, advertising or SEO scripts. Our public marketing pages still do.

• End meeting works on record-only meetings. A meeting with no agenda has no agenda rail, and the page's timer tried to update a clock that was not there. That error stopped everything after it, including the End meeting button, which rendered and did nothing. Reported by a customer with the screenshot that made the cause plain.
• A recording that loses its first piece is not marked saved. The first piece of a recording carries the header every player needs. If it could not be read after several retries, the recording used to be saved anyway and could not be transcribed. Your audio is now kept, nothing is deleted, and you are asked to retry the upload.
• Start is hidden from people who cannot start a meeting. Observers and inactive members saw a Start button the server then refused.
• Signals from the dashboard go on the meeting you pick. The form used to guess a meeting, usually the next recurring one, and ticking a Signal off could move it there. You now choose the meeting, and existing Signals are edited on their own meeting.
• Chart pickers stay the width of the panel. One seat with a long role stretched every Reports to and Escalates to menu off the edge. Roles are trimmed in the list with the full text on hover, and retired seats are no longer offered as someone to report to, while a link that already exists is kept.

• A person writes the rule. OTP never guesses which procedure answers which signal. You choose the entity or a wildcard, the level of drift that counts, the seat, and the procedure it runs. Rules are written on the drift page, and the priority and number pages each show the rules that already cover them so you can arm one where you are standing.
• Approving is not executing. A proposal that you approve is marked approved and nothing else happens. The work is a recommendation from the seat that owns the number, for a person to act on, and the page says so in both places rather than leaving you to find out.
• A procedure is allowed to say it does not cover this. An agent handed a situation its procedure does not address will otherwise comply anyway and produce a fluent paragraph that reads exactly like real work. It can now decline, and a decline switches that rule off rather than re-running it: the configuration is wrong somewhere, and paying to be told the same thing every sweep is not a fix. You meet it on the rule, with the seat's own sentence attached, not as an interruption.
• Three gates stand in front of anything that spends. It is off by default for your organization, because an armed rule spends your wallet with nobody in the loop at that moment. At the moment of firing it demands a funded wallet and an active key. And the daily cap on autonomous runs is genuinely shared, so on a day drift has fired your scheduled runs are throttled too. One wallet, one budget.
• Nothing dispatches at three in the morning. Detection still runs on its own schedule; only the dispatch waits for working hours. That needed a clock, so a human seat on your chart now carries a timezone alongside contact details. Empty means not stated and is never guessed, because a wrong timezone quietly schedules somebody's small hours and looks filled in on screen.
• Who may decide is what the chart says now. Deciding on a proposal takes the seat that currently owns the drifting item, or organization-wide authority. Arming a rule is a stricter, owner-level act, because that is the thing that spends.

• We were taxing the thing that makes the product different. An agent seat only became startable in August, and it carried a per-seat fee from the day it did. Charging for the seat and then charging for the work is two tolls on one road.
• You pay for work done, not for seats held. An agent that sits there costs nothing. An agent that reads, writes and runs draws from the wallet, and the wallet has been metering that all along, so this is not a new meter.
• Humans are free on every tier, with no footnote. That is the claim the comparison pages now rest on, and it needed to be true on every plan before it was worth printing.
• Enterprise is unchanged. Still two hundred and fifty dollars a month for twenty five agent seats, then ten each.
• The per-seat checkout is gone rather than switched off, and two quiet defects went with it. The settings page showed one price while checkout selected a different one, and checkout counted seats on the chart while the hourly reconcile counted a different population, so an organization could buy twenty and be reset to one within the hour. One population, counted by one function at both ends, can no longer disagree with itself.

• One choke point, not a list of toggles. Every AI feature in OTP resolves its key through one function, and the switch is checked there. That is what makes the promise hold for features that do not exist yet.
• Off means off, not off unless you brought a key. The switch is read before your own key is, so bringing your own model does not route around your own policy.
• Switched off now says switched off. The first cut of this handed an empty key to the provider, so an organization that deliberately turned AI off got an authentication failure back. That reads as OTP is broken when the truth is you turned this off. A blocked request now answers plainly: nothing is broken and nothing is owed.
• Ollie will not hand you a draft it did not write. A member stopped by the switch is told so, rather than being given a placeholder that looks like Ollie produced it.

• Ranked, and every line names its own evidence. Nothing in the brief is an impression. Each item points at the priority, number or to-do it came from, so the meeting argues about the thing rather than about whether the thing is real.
• Scoped to the meeting's own team. A brief carries that team's records and no one else's, so one team's preparation cannot show another team's numbers.
• No model, no spend. It is computed from stored rows, which is also why it needs none of the machinery that stops an unattended job spending your money: it cannot spend any.
• An empty brief says all clear, and a failed read never looks like good news. Nothing to report and could not read are stored as different things on purpose.
• You can turn it off by itself. Meeting preparation is its own category in your email preferences, so switching it off leaves everything else you asked for alone.

• Nothing was ever really destroyed, and now you can act on that. The rows have always been kept. The bin is the missing half: a list, and a way back.
• You see what you would have seen anyway. Your own to-dos, your teams' issues, and a private issue or priority only when it is yours. Deleting something does not make it visible to people who could not see it before.
• Restoring takes the same permission as deleting. Each kind of record reuses its own existing rule rather than inventing a second, looser one for the bin.
• There is no empty button, by design. No route can clear it, so nobody can turn a recoverable mistake into an unrecoverable one.

• A late completion no longer manufactures an overdue item. A recurring to-do sitting ninety eight days overdue, completed, produced its next instance six weeks in the past, and clearing that one produced another. The queue was generating its own backlog. The next instance now lands in the future while staying on its original cadence, so a sixty day rhythm stays a sixty day rhythm rather than becoming sixty days from whenever you got to it. An overdue count that includes items nobody could have done teaches you to stop believing the overdue count.
• A brief that stored but never mailed is still mailable. A meeting whose brief was computed while sending was off could never be picked up again, so it could never send, ever. Briefs are now selected by what they still need. An existing one is reused rather than recomputed, because it is a snapshot taken at a stated time and the meeting page has been showing it all day.

• A record, not three stacked cards. The person and their seat sit in the header, the facts you glance at sit in a sidebar, and the rest is tabs. Employment status, job information and compensation append rather than overwrite, so what somebody's title or pay was in March survives the day it changed.
• Who may open a record is decided by the chart, not by a job title. Owners and anyone granted HR access can edit. The person themselves, and every seat above them on the reporting line, get view only. A role on its own grants nothing: an admin with no seat above you does not find a door there at all.
• A balance you can trace beats a balance you are handed. Grants are the only stored fact and they stack, each carrying an author and a reason, so the days remaining are worked out in front of you rather than kept in a column that drifts. Weekends and the days on your holiday calendar cost nobody any leave.
• The approver is already on your chart. A request routes up the reporting line you drew, and nobody approves their own leave, an owner included. Approving shows what the absence actually costs: the to-dos due in that window, the meetings, the seats and the numbers that person owns.
• Calendars and policies decide what a day is worth. Holiday calendars are org level and plural, seeded from templates for the countries you employ in, every day editable. Policy tiers carry PTO by length of service, twelve days for the first two years and seventeen from the third, and bereavement is counted per death rather than per year.
• An HR record is about a named person, so you switch it on deliberately. It sits in Labs, off by default, with an owner-side settings page under People and time off for calendars, policy tiers and the two subcontractor tiers. Nothing new appears in your sidebar: records open from the members list you already use.

• Rename sticks, and status is a control you can see. The title only saved on Enter, and almost nobody presses Enter in a title field, so almost every rename died the moment you clicked back onto the canvas. Blur commits now, and attaching a board to a meeting promotes it out of draft by itself.
• Attach from whichever side you are standing on. Whiteboard is the first entry in the in-meeting Tools palette: it creates the meeting's board if it has none, opens the one it has otherwise, and attaches it either way, so a board started mid-meeting is never orphaned. From the board, Attach to meeting picks a recent one and detach undoes it.
• A board that cannot leave is a board nobody trusts. Download the page as a PNG or a one-page PDF, or send a link. The link serves a stored snapshot rather than an open door into your live canvas, and the viewer gives an email to see it. Re-sharing swaps the picture and keeps the link you already sent working.
• Photograph the wall, or just talk it through. Start a board from a picture, an iPhone photo or a single PDF page included, and from a spoken explanation. Ollie returns either a faithful diagram, with arrows bound to the shapes they point at, or distilled notes, and importing into an open board lands the result in its own frame.
• It transcribes the room, it does not improve on it. The reading is told to write down what is on the wall: no summarising, no additions, no tidied grammar, and a best reading with a question mark where the handwriting defeats it. A digitised board that quietly rewords the room is a board nobody in the room recognises.
• Ollie comments on the board, pinned to the note it means. Ask for a read and you also get up to five comments, each attached to one sticky: the question a note leaves open, what it quietly depends on, two notes that contradict each other, a decision hiding inside a statement. Never a restatement, never praise, and nothing to say is a fine answer.

• The bowl rings once, exactly when the section is over. It sounds as the clock crosses zero, not on a loop and not on every tick, and the warmup borrows it to mark the end of the minute of breathing.
• Muting is yours, not the room's. The bell in the rail turns sound off and on for you alone, and is remembered in that browser, so one person keeping their office quiet does not silence anybody else in the meeting.
• No files to load, and nothing plays out of nowhere. Both sounds are built in the browser at the moment they play. Starting the meeting is the gesture that wakes the audio, so a bowl due twenty minutes later can still ring, and a browser that blocks sound is simply silent.

• The two halves are joined before anybody tries to read them. A continuation segment with no header of its own is attached to the segment that carries one, and the pair transcribes in a single pass. A ninety-seven minute meeting that had been stuck for a week came back whole.
• A failure says what failed. The queue used to record the words transcription failed, whatever had gone wrong. It now keeps the provider's own error, which is the difference between a mystery and a fix.
• Ending a meeting is not destructive, so it stops asking twice. End used to arm itself for six seconds and wait for a second click, which nobody makes, because nobody reads a button they have just pressed. One click ends it, and a completed meeting can be reopened. Delete keeps its confirmation.

• The control appears for the people who may use it. An author can remove their own Signal and a facilitator can remove any of them. Deleting asks for a second press on the button itself, and a refusal from the server is shown to you rather than swallowed.
• The dashboard and the meeting page now agree. Permission to delete a Signal you wrote already existed in the API. The daily view was hiding it behind the facilitator gate anyway, so an author was locked out of their own mistake. Both surfaces read the same rule.

• Everyone on the same board at the same time. Live cursors and presence avatars, changes saved as you go, undo and redo. No screenshare, no one person driving while nine people describe what to move.
• It belongs to the meeting. Open a meeting, hit Actions, Whiteboard: start a fresh board named after the meeting, or attach a board you already have. The board is on the meeting from then on, which is the part a general purpose canvas tool can never do.
• Ollie reads the board. Ask for a summary and Ollie tells you what the board actually says. Ask for to-dos and it proposes them from your stickies, with owners, for a human to ratify. Nothing is created behind your back.
• Start from a sentence. Create with Ollie seeds a board from a prompt, so a blank canvas is not the thing standing between you and the session.
• A real toolset, not a demo. Stickies in thirteen colours with type-to-add and paste-a-list, twenty shapes, straight, curved and elbow connectors, freehand, and images dropped straight onto the canvas.
• On by default, yours to switch off. Every organization has it, existing and new, no backfill and nothing to enable. It is a beta and the toolset will grow fast, so if your team is not ready, switch it off in Labs.

• Sent when the work becomes real. Generating a read is a draft somebody is still editing. Accepting it is the approval, and the moment the to-dos exist, so that is when it mails.
• The whole read, not a clipping. The recap trims it to stay a recap. In this email the read is the point, so it travels intact.
• Once per meeting. Two people accepting at the same time cannot send the room two copies.
• A read, not a verdict. The email names how to challenge it, the same as everywhere else Ollie has an opinion.

• Pending looks like pending. The recap says Ollie is transcribing it now, rather than claiming no transcript or recording was attached.
• Failure is stated, not hidden. A transcription that gave up after repeated attempts used to render as silence. It now shows as a failure you can see.
• A second recording in flight is visible. The transcribing banner no longer disappears just because an earlier transcript already landed.

• Finalize is no longer serial. Chunks are fetched in parallel with byte order preserved exactly.
• Cleanup happens after you are served. Deleting the working files moved off the path you are waiting on, and cannot leave a recording half finished or resumable.
• Download while it saves. The file is available without waiting for the last step to finish.

• Same behaviour on both surfaces. Your own support requests and the full ticket view both take images on a note now.
• Paste works where you are looking. Pasting an image lands in the form you are actually using instead of whichever one was created first.
• A failed upload retries by itself. An image that does not make it stays staged with its error, so saving again retries only the upload and never duplicates your note.

• Share to network tells the truth. Your edits are live for your own team the moment you save them. That button only changes the copy other organizations can see. An owner removing somebody who had left was being asked to publish before the change counted.

• A note to whoever is steering, not a scoreboard. By default the score reaches the person who ran the meeting and the coach behind them. Attendees are off until an owner turns them on, because a judgement about an hour should reach the people who can change the next one before it reaches everybody.
• Written to be acted on, not filed. Each score comes with what to do differently, in plain English, short enough to act on the same day.
• Your coach sees what you see. A coach gets the same score, with one line explaining why it arrived. No separate report, no version of the truth the client never saw.
• It measures the hour, not the mood. Issues closed rather than carried. Decisions that became owned, dated commitments. Numbers before opinions. Quarterly priorities actually inspected. Off-track is reported, not punished.
• Still yours to switch on. Off by default, opt in from Labs, and it withholds the number entirely when too little of the meeting was captured to be fair.

• One list, not three. The rule lives in a single place now, so these seats cannot silently disagree again.
• Consistent with the rest of OTP. Visionary and integrator already counted as org-wide on the Issues board and in the invite ranking. Member management was the odd one out.

• A rubric, not a vibe. Issues closed rather than carried. Decisions that became owned, dated commitments. Numbers reviewed before opinions. Quarterly priorities actually inspected. Off-track is reported, not punished.
• Coaching in the email. Each score comes with what to change next time, written so you can paste it to the team.
• It knows when it cannot know. Below half the evidence, the score is withheld and the page names what was not captured. A 0.0 on a fifth of the meeting is not a finding.
• See what did not score. The page lists every meeting that produced no number and why, and an owner can fix the list in place (exclude a standup, mark a meeting as not one of yours).
• Off by default, and not in the sidebar. A judgement about how your team ran an hour is something an owner opts into, from Labs, on purpose.

• Manage, then Members, Teams, Portfolios. The scale ladder reads left to right: people, teams, organizations.
• Asked once, at the right moment. The default is one company, so nothing changes for the ordinary signup.
• Setup staff are not stakeholders. When our team builds a portfolio for you, we no longer end up seated on it, and an owner can leave a portfolio they no longer want.

• Append-only. A note is a record, not a box that gets edited over.
• Three visibilities per note. Private (no admin override), owner, or team.
• On the priority page and on the meeting. One shared view, wherever the priority shows up.

• Open means start. The page showed a Start button to everyone who could read the meeting and then refused the click unless they were on the team. Same rule for ending.
• Personal is an answer. Picking no team records that choice, so the list stops asking.
• Locked, not dead. A future occurrence keeps its body locked and its controls working.

• Library tiles. Thumbnails render the image; the written description opens on click; a dead link fixed.
• Support replies. Multi-line, auto-growing reply field.
• Honest boot messages. Two failure notices rewritten to say what actually happens.

• One door, wherever you are. "Hire your first agent" now appears in the first-run checklist, on an empty org chart beside "Add your first person", and as a visible button on the chart header.
• Every seat has a work record. An agent logs what it did against its own seat, so the chart shows work rather than intent.
• An agent can claim a seat you already drew. You sketch the seat like any other; the agent claims it and starts filing work against it.
• No paid gate. The connection is on by default for every organization, free included, and stays switchable from Labs.
• We went first. Ten of our own agents claimed seats and started logging work this week, which is how the rough edges got found before you hit them.

• Your phone is now an extension of the meeting. Photograph the board without stopping the recording or moving to a laptop.
• Ollie describes what it sees. Every image gets a written description, which is what makes it findable months later.
• Your words win. If you edit a description by hand, nothing overwrites it.

• Search the contents. The words Ollie wrote about each image are what you search, so "the pricing whiteboard" finds the pricing whiteboard.
• It respects who can see what. Files inherit the visibility of whatever they hang off. A photo on a private issue stays private to its owner, admins included. That rule is the feature, not a footnote.

• Carry, do not copy. One file, linked in a second place.
• Every existing meeting attachment was backfilled, so this works on the meetings you already ran, not only new ones.

• Named attendees can see their own meeting. Previously a direct report could be the subject of a 1-on-1 and never see it.
• Agendas arrive pre-seeded, so the meeting starts from something rather than a blank page.
• Open items carry forward to the next one instead of being retyped.

• Three levels, watch, at risk and slipping, each with the reason attached.
• Every call cites its evidence. If you disagree you can see exactly which fact produced it.

• Invite link, not a public listing yet. The app is registered and usable now; the public directory listing comes after the shakedown.
• Per-key budgets, so one noisy automation cannot spend the whole organization's allowance.

• Read-only and public by design. No org data, no writes, no pricing or roadmap answers.
• Separate from your own connector. The authenticated one that reaches your organization's data is unchanged.

• A receipt on every line. No claim without the record behind it.
• A released to-do is a change, not a kept promise. The report will not launder one into the other.
• The person at the top goes first. Structurally, not by convention.

• One switch per thing we send. Meeting recaps, work reminders, the getting-started course, In the Swamp, and the programs still being built, which appear now marked "Coming" so you can decline them before the first one ever arrives.
• Pause everything, without remembering to undo it. One week, one month, or until you turn it back on. A timed pause expires by itself.
• The mail you cannot switch off is listed too, with the reason. Invitations, account and security messages, and anything you specifically asked for. Nothing reaches you that is not on this page.
• It opens straight from an email. No sign-in needed, which matters if you are on our list without an OTP account.
• Every change is on the record. Settings, My data now shows each one, when it happened, and whether it came from the page or from a link in an email.

• No opt-in step. Connect a calendar and Ollie starts showing up. Nothing to be enrolled in first.
• Per person, not per organization. Each member decides whether their own meetings get an agent, from that same page.
• Billed from your workspace wallet, the same as sending Ollie to a call by hand.

• Enough to get through week one. Send Ollie to a meeting, work what comes out of it. That is the whole starter path.
• The rest reveals itself. Quarterly Priorities, KPIs, Issues and Notepad appear after a first completed meeting. Team Chart, Members and Teams appear when someone else joins.
• Nothing is taken away. This only governs what a genuinely new organization sees. An established workspace is untouched, and a revealed item never hides again.

• One field. "Who was in the room" should not cost more than typing their name. Email is optional and can come later.
• In place. The person is created and selected without a page change, so you keep your place in the meeting.

• "What needs your attention" stays put. It used to be pinned to the height of the column beside it, so a long left column pushed it far down the page and out of sight. It now holds its own position.
• "Just mine" is the default view. The dashboard opens on your Quarterly Priorities, KPIs and Issues. Switching to everything on your teams is still one click, and any link or bookmark that already specified a scope keeps working.
• Upcoming meetings shows today and tomorrow. The rest is one click away instead of filling the tile.

• What went wrong. New timeouts meant to end no-show calls were tuned too tightly and fired during a real conversation.
• What we did. Both timeouts are disabled, so the request is identical to the one we sent before the change. If your meeting was cut short that afternoon, this is why, and it cannot happen the same way again.
• Coming back carefully. Leaving an empty call is still worth doing, so it will return with thresholds proven against real meetings first.

• Speaker assignment, attendee pickers and owner lists all collapse to one row per human.
• Seats are not lost. Holding two seats still reads correctly on the chart; it just stops turning into two of you in every dropdown.

• Formatting renders. Lists look like lists, emphasis looks like emphasis. Previously a reply arrived as one wall of text with the Markdown still in it.
• Every destination is a link. Pages, settings, billing: if Ollie names somewhere you need to go, it is clickable. External links open in a new tab.
• Answers end on the next step. When something cannot happen until you do something, he says what to do rather than only what is wrong.

• Scheduling by asking. "Run this every morning at 10" is now a sentence Ollie can act on. He reads your real processes first, so he schedules the one you meant instead of guessing at a name.
• The cadence is on the approval. Nothing is scheduled until you say yes, and what you are saying yes to reads as "every day at 10:00 (America/New_York)" rather than a bare "create a schedule". Approving a schedule approves every run it will make, so the card shows the schedule.
• Run it once first. Ollie can run a process on demand, which is usually the better order: see the output, then decide whether it earns a place on the calendar.
• Pause beats delete. Ask him to stop a schedule and he pauses it, keeping the setup so you can turn it back on in one click.
• He states the limits. Cadences are hourly, daily and weekly, so "every weekday" is not something OTP can express exactly. Ollie now says that plainly instead of quietly rounding it to daily.
• New in the MCP server. The same five capabilities are available to any connected agent: list your processes, schedule one, change or remove a schedule, and run one now.

• Stale connections correct themselves. A connection whose authorization window closed used to sit there saying "pending" forever, because nothing ever re-checked it. Now any connection anyone looks at is reconciled against the provider first.
• You get the actual reason. "Connection initiation did not complete within 10 minutes" means the consent screen was never finished. That sentence now reaches you, on the integrations page and from Ollie, instead of being discarded.
• Ollie stops saying he does not know. A connection that is expired, revoked or still authorizing loads no tools, which he used to read as the connection not existing. He now sees every connection and its state, and says what is actually wrong.
• He can look things up. Ollie searches the catalog of over a thousand connectable systems, and reads Composio's own documentation for scopes, setup and available actions rather than recalling them.
• Dead ends become tickets. When you ask for something he genuinely cannot do, he says so and offers to raise a support ticket instead of leaving you with a no.

• Connect once. Google Calendar, from Settings, Meeting presence. Ollie watches for upcoming meetings that have a video link and joins them.
• He knows whose bot he is. The participant list shows Ollie named for your organization, so when two OTP companies meet each other it is obvious which side each bot belongs to.
• It lands on the right meeting. A captured call is matched to a meeting you already scheduled by its video link and start time, so the recording, transcript and follow-ups attach to that record instead of creating a stray one.
• Captured calls find their team. A meeting captured by the person who connected the calendar is routed to their team, and anything unassigned is surfaced rather than silently filed.

• Read it. Ten laws, the limits on Ollie, what OTP refuses to become, and how the document itself gets amended, published verbatim from the governing document in our repository at /constitution.
• Challenge any insight. Every Ollie Insight now carries a Challenge action. A challenge states its grounds, targets the whole read or one cited receipt, and travels with the insight even when it is regenerated. Anyone can respond; only the person who raised it can close it. Rank cannot make it go away.
• See what OTP holds about you. Settings, My data shows your identity record, what your organization holds that involves you, and who has viewed your meeting transcripts.
• Commitments end on purpose. A to-do is kept or released on the record instead of fading, and repeated due-date moves show as quiet receipts.
• Leadership goes first. When last week's words are reconciled against the week's deeds, the top of the chart is chased as loudly as everyone else. Accountability that only flows downward is surveillance, not integrity.
• Evidence before opinion. Ollie asserts nothing it cannot cite: every insight arrives signed, with clickable receipts, and can answer why you are seeing it.
• Your mailbox is yours. Mail and calendar integrations now connect per person, as that person, and work only in their own Ollie chat. One admin can no longer bind a mailbox for a whole organization.

• Capture a real account, not a template. A Blueprint records how a company that already runs well actually works, then applies it to another organization in one click.
• Seven things travel. Language, meetings, KPIs, structure and seats, Ollie's voice, quarterly priorities, and processes. You choose what goes in, and every asset is listed by name before anything is applied.
• Configuration travels, your data never does. A Blueprint carries the seat, never the person sitting in it, so nothing private moves between organizations.
• Versions that stay honest. Improve the source once and push a new version to every linked organization. The update preview tells the truth per asset (new, unchanged, a safe update, or something the client changed themselves), and every apply can be undone.
• Published Blueprints become templates. A new organization starts already inside your method instead of onboarding to an empty tool.
• Now live. Blueprints graduated out of Labs, no switch to flip.

• Ask in plain language. Streaming answers, three starter prompts, Enter to send.
• Watch it work. When Ollie reaches into a real source it shows the step live ("reading your quarterly priorities") and links to what it found, so nothing happens off screen.
• Grounded in your org. This is Ollie reading your own data, not a generic chatbot.
• Early access. A Labs preview with one live source wired so far and more coming. Turn it on under Settings, Labs.

• Auto-generate on confirm. The insight starts writing the moment you confirm the transcript, so the recap is ready when you are. Labs, and only while credit remains; a plain refresh never re-charges.
• Give Ollie context first. Add a short organizer note before the meeting and Ollie folds it into what it writes.
• Out of your way at zero balance. Nothing auto-charges, and the manual button is still there when you want it.

• An owner every time. Every teammate is always assignable, so no action item is left ownerless.
• Priority and a deadline. Mark what matters and when it is due, right on the item.
• Smarter owner-guess. Ollie proposes who owns each action, and you correct it in a tap.
• Add and batch. Add items by hand, and select many at once to assign or update them together.

• Pause any time. Stop capturing for a private moment or a break.
• Resume into the same recording. No new file and no lost context; it stays one meeting.

• Private by design. Only admins and the creator can open a restricted meeting.
• Same meeting, narrower door. Recording, transcript, and insights all work as usual.

• List pages reflow. Priorities, to-dos, and issues stack into one clean column instead of a wide table you have to drag.
• Your KPIs stay readable. The title column pins in place while the numbers scroll, and the grid is tappable.
• Switch orgs on a phone. The org switcher and settings navigation work at phone width.
• One layout, everywhere. A single mobile breakpoint across the app, not a patchwork.

• Start from your files. Drop in what you already have and Ollie reads it into a working OTP.
• Review before it lands. Ollie stages what it found; you keep what fits with a checkbox and build in one click.
• Straight to your chart. New users land on their Team Chart, not a settings maze.
• Reusable later. The same import lives under Settings for whenever you want to give Ollie more.

• One grammar everywhere. Priorities, KPIs, and issues share the same row: bold title, one muted line of meta, a small status pill. Your eye learns it once and reads the whole page with it.
• Milestones fold away. Each priority shows its progress count; expand it only when you want the checklist. The sub-task sprawl moved to the priority's own page where it belongs.
• KPIs at half height. Slim inline entry that saves on Enter, trends in a fixed lane, numbers lined up on the right.
• Issues you can actually edit. Editing now covers the description, not just the title, and a truncated description expands right in place when you click it.
• Start a meeting from the dashboard. New meeting sits next to the Meeting panel and opens the same flow you already know.
• Quieter by default. Headlines and Insights start collapsed and remember your choice.

• The whole org in one grid. Members down the side, sidebar items across the top, checkboxes in between. No per-person settings safari.
• Copy a setup. Got one seat configured right? Copy it to the next hire in one click.
• Templates for repeat roles. Save a permission set with a name and apply it to any member. Build it once, use it every hire.
• Admin wins, safely. Items you turn off stay off even if the member customizes their own sidebar, and every page behind an item stays gated by role regardless.

• Two benches, one meeting. Voice A and Voice B read the same transcript, side by side, so the difference is audible instead of theoretical.
• See where the voice did the work. Highlights mark the phrases an influence shaped, and every to-do, issue, and headline can carry a tag naming what framed it.
• Tune the influences inline. Click any influence to open its principles: remove what does not fit, add what your company believes, then run it again.
• Save the winner. One click writes the voice to your org or any team you lead, and every insight after every meeting arrives in it.
• Early access. Turn on Ollie Lab under Settings, Labs. Each run meters like other AI assist.

• Twelve thinkers, twelve fingerprints. Every library influence carries a rhetorical fingerprint written from their published voice, so the difference between them is audible within two sentences.
• Catchphrases banned. If a phrase would look at home on a book cover or a poster, your Ollie will not say it. Voice lives in structure, not slogans.
• Your own influences too. Add any name and Ollie reconstructs the published voice from structure and register, carried honestly.
• Never impersonation. Composed, never copied: the voice never speaks as anyone, never borrows their words, never claims their blessing.

• Door one: the full experience. Chart, scoreboard, priorities, and your weekly meeting in OTP. Offered, never required.
• Door two: bring the meeting you already ran. Paste the transcript from Zoom, Teams, Meet, Fireflies, or plain notes, and Ollie pays you back in minutes. Zero change to how you work.
• Door three: no meeting at all. Connect the numbers you already track and your scorecard fills itself in, timestamped.
• $5 of Ollie credit at signup. Every new organization starts funded, no card required, so the first Ollie runs cost nothing.

• The real pipeline, not a demo. The taste runs the same reading Ollie does inside the product, on your actual text.
• Blurred first, stored never. Emails, phones, and secrets are blurred before the model reads, the response tells you how many, and nothing is kept.
• Worth sharing. Send a skeptical friend to the homepage and let them paste their own meeting. It argues better than we do.

• The proof. Britain, summer 1940: the facts did not change, the voice did, and Murrow named it: he mobilized the English language and sent it into battle.
• The constraint. You can be in one room at a time, and a summary is a voice with the blood drained out.
• The turn. A voice can now be composed, one per room, yours to overrule, and motivation compounds exactly like knowledge.

• An Ollie per scope. Shape one Ollie for your whole organization and a distinct one for each team, so a leadership Ollie and a sales Ollie can sound and think differently.
• Gurus you pick. Add the voices you want in the room, real thinkers you name or a guru you describe, and Ollie brings their lens to your insights and answers. Build your own from a short description.
• Personality and governance. Set the tone, and set the boundaries: what Ollie may weigh in on and what it leaves alone.
• Ask Ollie anywhere. In-app help now answers in Ollie’s voice, from one canonical place across the app.
• Free to try. Your first Ollie Insights generation on a meeting is on the house.

• A Warm-up that opens the room. The meeting starts with a short, calm arrival: a breathing moment and a "leave it at the door" beat, so people are present before the work begins.
• A clearer flow. Warm-up → Personal Updates → Ollie Insight → Signals → The Board → Quarterly Priorities → To-Dos → Issues → Wrap up. Signals moved up so momentum comes early.
• Plainer names, un-numbered. Every section reads in plain language and the numbers are gone, so the agenda reads as a flow, not a checklist.

• An insight, not a summary. Ollie leads with the light: what went well, momentum to build on, and one or two warm nudges toward a more productive next meeting.
• Carried forward. The insight appears at the top of your next meeting, right after Personal Updates, so the team reads it to open.
• Patterns across weeks. Each insight looks back at recent ones to name a trend that is building, not just this week in isolation.
• Ollie-made to-dos are marked. Any to-do Ollie generates from a meeting carries an Ollie badge, so it is clear what came from the transcript.
• Early access. Turn on AI meeting insights under Settings, Labs. Generating an insight draws from your wallet's AI credits.

• Your section names, live. A format you build in the format editor now flows straight into the meeting: each section shows the title, timebox, and order you set, not a fixed template.
• A free-form discussion section. Add an open notes section to any custom format for the conversation that does not fit a box.
• Defaults stay clean. The built-in Delta Meeting is unchanged; custom wording only ever appears because your team typed it. Turn on custom formats under Settings, Labs.

• Cassette controls. Previous, play/pause, and next, right under the digital clock.
• No more scrolling. The agenda list stays fully in view.
• One action to start. Pressing Start meeting now starts the clock automatically.

• The path, in order. Build your team, add a seat, set a KPI, and run your first meeting, each step explained in a line.
• Self-checking. Steps tick off as you complete them, and the whole checklist hides once your setup is done.
• No dead ends. Every step links straight to the page that finishes it.

• Wallet top-up fixed. Adding credits now opens checkout reliably every time.
• Invites tell the truth. When an invitation email fails to send, every invite and resend surface now shows it and offers a recovery path, instead of a false "Invite sent."
• Friendlier notices. In-app toasts replace the browser's native alert pop-ups throughout the app.

• Insert, don’t type. Pick a KPI from a menu and it drops in as a link; add +, -, ×, ÷ and parentheses to shape the math.
• Live preview. The formula resolves as you build, so you see the result before you save.
• Computed, not editable. A formula KPI carries an “fx” marker and updates itself when its inputs change, on your dashboard, the Scoreboard, and inside a meeting, so its value can’t be overwritten by hand.

• Private to you. Your notepad is yours alone, not shared with the team.
• Always saved. Notes save as you type, so nothing is lost.
• Formatting when you want it. A light toolbar and a preview for headings, lists, and emphasis.

• Short-term vs long-term. Tag each issue by horizon and filter the daily Issues panel to just what you want to work now.
• Filters that stick. Your horizon filter is remembered the next time you open the panel.
• Merge duplicates. Fold repeat issues into one super-issue that preserves all of their notes and details.

• Import from the UI. Paste or upload a transcript from a past meeting, no setup required.
• Review before it saves. A pre-ingestion review step and an immutable edit audit trail keep the record honest.
• Ollie reads it. The imported meeting shows on your meetings page and feeds insights, to-dos, and issues.

• + Add KPI on the Scoreboard. A button in the header and on the empty state, available to KPI admins, opens a create panel without leaving the page.
• Pick the owner inline. Choose the owning seat from a dropdown in the panel, so a new KPI lands on the right seat immediately.
• Team chart unchanged. You can still add a KPI from any seat on the Team chart, exactly as before.

• Private by design. A Shadow Rock is visible only to the person who owns it. No one else in the organization can see it, not managers, not admins, not the owner of the org.
• One checkbox. Mark any Rock private from the Rock form or the SMART Rock Planner. It shows a Private badge wherever you see your own Rocks.
• Hidden everywhere else. Shadow Rocks are filtered out of every other member’s Rock Review, dashboard, profile, rock list, and meeting view, so a sensitive priority stays yours alone.

• Automatic on sign-up. A matching pending invite is accepted the instant the account is created, so invited teammates land inside your org with their seat, team, and access already set, instead of in an empty account.
• Meeting runner fix. The meeting runner no longer renders two sections at once.

• Super-metrics. Roll one KPI per member org up into a single number, with per-org weights, so "leads this week" across a dozen locations becomes one line on the portfolio scorecard.
• One meeting across the group. Start a portfolio meeting that reports the rolled-up scorecard, with attendees drawn from your member organizations.
• An organization template. Define your standard once, the starter KPIs, the sidebar items every location shows, and the AI-key policy, and every member org inherits it. Lock it to enforce one brand standard everywhere.
• Bring in a location by email. Invite a champion; when they accept, a brand-new organization is created from your template, owned by them, and linked into the portfolio, ready on day one.
• Member org chart. See every member organization as a card with its people, agents, and KPI counts, and open a side panel for the details.
• Early access. Enable "Portfolio" in Settings, Labs. Member organizations keep their own data; only the metrics you choose roll up.

• One hosted URL. Generate a connection from Settings, API and paste it into Claude (Settings, Connectors, Add custom connector) or any MCP client. No npx, no config file.
• Your org, scoped. Your agent gets your real chart, scorecard, rocks, to-dos, and the coordination graph as tools, limited to your organization.
• Early access. Enable "Remote MCP connection" in Settings, Labs. A paid plan is required.

• 1,000+ apps. The complete catalog, searchable by name, so you can find your stack fast.
• Connect from Settings. Link a tool in a few clicks under Settings, Integrations.

• Your sections, your agenda. Build a format with the parts your meeting actually needs, then run it live in OTP.
• Start from a template. Import any of the 180+ library agendas as a starting point.
• Early access. Enable "Custom meeting formats" in Settings, Labs.

• Groups. Add, rename, delete, and reorder KPI groups, and move any metric into the group where it belongs.
• Full inline edit. Every KPI row opens a complete edit panel (name, target, owner, group) right on the scoreboard.

• Opt in per organization. Admins choose which early-access features to switch on; nothing changes for anyone until you flip the toggle.
• See what's cooking. Each Lab explains what it does and why you would want it on now.

• End at the bottom too. The End meeting button now lives at the foot of the meeting, where you naturally wrap up, as well as the header.
• One-hour auto-end. A meeting still running an hour after it started ends itself, so nothing is left open forever. Press Extend to add another hour when you need it. Everything entered is kept.
• Future recurring meetings stay locked. Next week's occurrence does not open until its date arrives (or you start it on purpose), so your check-ins, headlines, to-dos, and issues land on the current meeting, not a future one.

• Filter dropdown. Switch between All, Agents, and Humans from one tidy control.
• Actions dropdown. Run Leadership Meeting, People Review, Invite member, Import CSV, and Pending invites are grouped together instead of crowding the header.
• Zoom. Use the on-screen controls (bottom-right of the chart) or Cmd/Ctrl + scroll to zoom in and out, then click the percentage to reset.

• Archive. Hide a priority from the board while keeping it on record, recoverable later.
• Delete. Remove a priority you no longer need, right from the card.

Every meeting worth running has a shape. We wrote down more than 180 of them.

• Every major operating system. The classic entrepreneurial operating system, Scaling Up, Pinnacle, Accelerate (AOS), Metronomics, OKR, 4DX, Holacracy, Hoshin Kanri and Lean, V2MOM and OGSM, and Balanced Scorecard, each with its full meeting cadence from daily huddle to annual planning, plus signature sessions like the 4DX WIG Session and the Holacracy Tactical and Governance meetings.
• And every everyday meeting. Agile and Scrum ceremonies, 1:1s and people meetings, team operations, retrospectives, customer and external, and innovation and facilitation.
• Real agendas, not stubs. Each template has a timeboxed agenda, who should attend, duration and cadence, facilitator tips, and common mistakes to avoid.
• Search and filter. Find a template by name, methodology, or what you are trying to do. Filter by operating system or meeting type.
• Free to take or run. Download any template as markdown, print it, or open it and run the meeting live in OrgTP. Every template is public, no sign-in needed.
• Browse the library at /templates.

The whole operating system, behind a closed door, enforced everywhere.

• One switch. Settings, Configuration has a Private mode toggle (admins only). Flip it on and your organization is excluded from every public and cross-organization surface at once.
• Nothing leaks, nothing is borrowed. Your org chart, OOS, scorecard, and claims stay out of browse, search, the graph, compare, and the directories, and they are never used to generate recommendations for anyone else.
• Your team is unaffected. Everyone inside your organization keeps full access to your own data through the dashboard. Private mode only closes the outside door.
• Reversible. Turn it off any time and your organization rejoins the network.

• Searchable guide. Open the User Guide (or the Help panel), type what you are looking for, and matching sections surface instantly with the terms highlighted. Shareable: /guide?q=scorecard deep-links a search.
• Searchable articles. The /blog library filters live by title, summary, and tag.
• Connect Your Agent. A focused onboarding page with a copy-paste AI prompt that generates your first OOS in about 60 seconds, plus the one-line Claude Code install. Linked from the guide.

Hit Customize on the Daily dashboard.

• Per-row shapes. Each row picks its own pattern: full width, 1/3 + 2/3, 2/3 + 1/3, half + half, or thirds. Mix them however you work.
• Drag anywhere. Tiles move between cells and rows, not just within a column. Removing a row hands its tiles to the row above -- nothing disappears.
• Bigger type. Six sizes up to 3XL for wall screens and tired eyes, with tiles that wrap gracefully instead of clipping.
• Yours alone. Layout and font save per member automatically; teammates each keep their own.

Rocks tell you the destination. Milestones tell you whether you are on the road.

• Add milestones on the Daily dashboard. Open a Rock card: title + due date, done. A progress chip (2/5) sits on the Rock and turns green at 100%.
• Due dates that mean it. A past-due, unchecked milestone shows its date in red.
• Assign to-dos to a milestone. Each milestone takes its own to-dos with owners, picked from the same people list as delegation. They show up in the owner's queue like any other to-do.
• Check off in the meeting. The Rock Review section of your weekly meeting lists each Rock's milestones with live checkboxes.

• Paperclip on every to-do on the Daily dashboard and your to-do queue: attach, list, download, remove.
• Carry, don't re-upload. From a to-do's attachment, send the same file to an Issue or a Rock with the arrow buttons. One file, linked wherever the work lives.
• Issues and Rocks too. Attach files directly to either, from their cards on the Daily dashboard.
• Sensible limits. Up to 5 MB per file; removing the last link removes the file.

Everything in Free, plus a closed door.

• Nothing published. Your humans, agents, seats, OOS, KPIs, and meeting data stay out of every public surface: browse, search, compare, and the intelligence graph.
• Nothing borrowed. Private org data is excluded from network intelligence -- no cross-org learnings, patterns, or recommendations draw on it.
• Invitation only. You decide exactly who sees inside your organization.
• Founding-member pricing. $99/month. See the pricing page or speak with our team to get set up.

Navigation that stays put while you work.

• Everything in one rail. Daily, Meetings, Team chart, KPIs, To-Dos, and Issues up top; Operating Plan, CEO view, Members, Teams, Docs, Workspaces, Inquiries, Publisher, and API keys appear based on your role; Tickets and the Guide anchor the bottom.
• Collapse it. The chevron at the bottom shrinks the rail to icons. Your preference is saved to your profile and follows you between sessions.
• A focused top bar. Signed in, you see Ask AI, What's New, Help, alerts, Publish, and your account -- nothing else competing for the space.

Learn the system by asking it.

• Ask anything about OTP. "How do I create a Strategy Reset meeting?" "Where do I see my team's KPIs?" Answers arrive in seconds and point you to the right page.
• Grounded, not improvised. Answers come from the user guide, the route map, and recent product updates. When something is outside its knowledge, it says so and offers the ticket desk.
• Built to grow. This is step one -- answers about your own org's data are on the roadmap.

One dashboard, arranged per person.

• Drag to reorder. In Customize mode, grab any tile in the left or right column and put it where you want it.
• Hide what you don't use. Every tile except the cascading message can be hidden -- and brought back any time from the same mode.
• Three font sizes. S, M, or L, applied to the whole dashboard.
• Saves itself. Changes persist to your member profile as you make them. No save button, no setup.

Create one from the Meetings page: New Meeting, type "Strategy Reset."

• Nine guided sections with timeboxes, from problem definition (15 min) through strategic options (45 min) and the debate (60 min) to written commitment and success criteria.
• The deliverable test. The meeting is not done until leadership can answer four questions with one voice: the problem, the decision, the 90-day priority, and the actions.
• Notes autosave into the meeting record; Assign Ownership creates tracked to-dos with owners and due dates.
• Bonus for every meeting type: the title field is now optional -- leave it blank and OTP names the meeting from its type and team, like "Leadership Meeting -- Leadership Team."

• The megaphone. A red badge appears when updates have shipped since your last look. Open it for the latest entries; the full history lives on the What's New page.
• Help and Support. The (?) in the nav opens a panel with Raise a Ticket, the User Guide, What's New, the operating-system + AI article library, and what's coming next.
• No more exits. Guide, What's New, blog, and Premium Support pages keep the sidebar and top bar when you are signed in. Visitors still get the full marketing pages at the same links.

Details and the full Standard-vs-Premium comparison at /premium-support, or find it in the Help panel.

Two halves of the same loop: see what moved, and unstick what didn't.

• The bell. Between Dashboard and Publish in the nav, with a red unread count. Opening it shows your latest alerts and marks them read.
• What lands there. A to-do assigned to you, a to-do you delegated being finished by its owner, and your to-do being closed out or verified by someone else.
• Browser notifications. Grant permission once (the bell asks on first click) and new alerts also fire native desktop notifications while an OTP tab is open.
• Close it out. On the Daily dashboard and /me/todos, every item you are waiting on others for now has a one-click disposition: done + verified in a single step, with the owner notified.

For when a whole scorecard group stops being part of your organization.

• Archive from the scoreboard. An admin-only action on every KPI row. Archived KPIs leave every default view: scoreboard, Daily, meeting scorecard snapshots, founder-dependency and accountability diagnostics, and publish-all.
• Never a delete. Every value ever recorded stays in the database. Archiving is a flag, not a removal.
• Show archived. A toggle next to the scoreboard view tabs reveals retired KPIs, badged and dimmed, with an unarchive action to bring one back.

Built so a first-time user can learn the system from inside the system.

• Everywhere you work. Daily dashboard panels, every agenda section of the meeting runner, to-dos, the scoreboard, issues, the org chart and People Review, Members and Teams, the Operating Plan and CEO view, Browse and the Intelligence Graph, and Settings.
• Plain language first. Each popup defines the term (Rock, KPI, IDS, OOS, seat) in one line, then what the panel shows, then what you can do there.
• Small touches. Escape closes the popup and focus returns to where you were.

A batch focused on making invitations feel real immediately, plus a cluster of fixes.

• Invited people appear as pending tiles. The moment you invite someone, they take a seat on the org chart marked pending, so the chart reflects your real team before anyone clicks accept.
• Onboarding can seat their work, not just yours. During setup you can now assign Quarterly Priorities and KPIs to the people you invited, so a Rock or a number can already point at its owner while the invite is still outstanding.
• Scoreboard shows the name, not the ID. For a seat nobody has claimed yet, the scorecard now reads the member name instead of a raw identifier.
• Signup is solid through the last step. The onboarding flow now loads the sign-in provider correctly, so the writes on the final wizard step no longer fail with a permissions error.
• Cleaner sign-out and invite form. Logging out now reloads and lands you on the sign-in page, and the invite form uses neutral placeholder text.
• Public pages no longer error when signed out. The Browse (Claims) and Intelligence Graph pages had a server error for logged-out visitors; both now load.

Everything you configure now lives in one place, reskinned to the light in-app look.

• One hub, one menu. A picture menu opens the whole set: Profile, Preferences, Notifications, Account, Integrations, Configuration, and API Keys.
• Profile got real structure. Personal, Metrics (work-style: MBTI, Kolbe, CliftonStrengths), Contact, and Social tabs.
• Billing reflects your actual AI team. A live agent count and a real per-agent cost model -- humans free, each active agent billable -- with the subscription and upgrade UI in place.
• API Keys folded in. The key list and MCP setup snippet now match the rest of the hub instead of the old dark header.

Layout and meeting-list polish aimed at the screen you open every morning.

• Full-width tile layout. Heavy tiles take a two-thirds main rail; light tiles and your Agents / OTP Insights sit in a one-third sidebar. Headlines and Quarterly Priorities now open by default.
• The full meeting cadence. The list now offers Weekly Leadership, Departmental, Quarterly, Annual, and One-on-One, with a (?) explainer for what each one is for.
• Recurring to-dos behave. A to-do you switch to recurring no longer vanishes from Daily, and you can set recurrence straight from the Daily quick-add.

Less visible than a new screen, but the foundation a coordination platform has to get right.

• Tickets stay inside your org. Reading and writing support tickets is strictly scoped to the caller organization.
• Script-context output is escaped. Any value rendered into an inline script is escaped, so content that happens to contain markup can never break out of the script. A lint rule now enforces it on every page.
• A real safety net on releases. The test suite and linting are now a blocking gate on every push, including an integration harness that specifically checks for cross-tenant access regressions.

Five changes, all aimed at making the weekly leadership meeting behave the way you expect when you are running it live.

• Rocks update live during the meeting. The Rock Review is where you set On Track / Off Track and mark Rocks complete, so those edits now render immediately. Previously the page showed a snapshot frozen at meeting start, so a change you just made looked like it had not saved (it had). Rocks are now always live; only the scorecard freezes.
• "Changed this meeting" note. When a Rock's status moves during the meeting, it shows what changed ("Off Track → On Track", "Completed this meeting") instead of just flipping with no record. A Rock you complete mid-meeting stays visible with the note rather than vanishing.
• Scorecards are team-clean. A meeting only shows the KPIs that belong to its team. A metric owned by another team can no longer surface on the wrong meeting's scorecard.
• Final numbers captured at end. Ending a meeting now re-captures the scorecard, so the completed record reflects the numbers as reviewed, not the ones from when the meeting started.
• Recurring delete asks first. Deleting a repeating meeting now offers "only this meeting", "this and all following", or "the entire series", and the series-ending choices actually end it, instead of the calendar quietly regenerating the next occurrence.

Shared meeting links also unfurl with a new branded preview card in Slack and other apps.

The import engine was built source-agnostic on purpose: it reconstructs your chart from the owner column that every clean export from your current tool carries, regardless of which tool produced it. Adding Bloom Growth was mostly teaching the parser Bloom's file names.

• New page at /import/bloom with Bloom-specific export steps (Bloom exports per meeting, as a ZIP you unzip). Same preview-then-commit flow, same mascot, same "rebuild the chart from who owns what" result.
• Bloom's quirky file names handled. Its To-Dos file is literally named "To-Dos (KPI (Metrics))", which naive detection would mistake for a scorecard. The parser now matches explicit module names first and treats the scorecard as the fallback, so To-Dos, Quarterly Priorities (Goals), and O&O (Issues) all land in the right place.
• Secondary by design. Ninety stays the headline; Bloom is a quieter second door. The two import pages cross-link, and the engine, endpoints, and commit logic are shared.

Note: like Ninety, Bloom's accountability chart is print-only on export, so OTP reconstructs it from owners rather than parsing a chart file. Bloom also has a REST API, which is a cleaner future path than CSVs; for now the drop-the-files flow keeps it consistent with Ninety.

The biggest thing keeping anyone on their current operating-system tool is the cost of switching. This release attacks that directly for Ninety.io, the most common place operating-system-style teams keep their Rocks, scorecard, and issues today.

The hard part, and the trick that beats it

We looked at exactly what Ninety lets you export. The five list modules come out clean as spreadsheets: Rocks, To-Dos, Issues, Headlines as XLSX and the Scorecard as CSV. But the one thing OTP cares about most, the Accountability Chart, has no structured export at all. Ninety only prints it to PDF, and the detailed PDF drops the seat descriptions. There is also no public API.

So we do not need Ninety's chart. Every clean export carries an owner name on each row. OTP reads those owner columns across your Rocks, KPIs, To-Dos, Issues, and Headlines and rebuilds your roster from who-owns-what. That is the demo moment: you drop five files and watch your chart come back, seat by seat, reconstructed from accountability rather than from a picture.

How it works

• Preview, no account, nothing stored. Drop your files at /import/ninety. They are parsed in memory and discarded. You immediately see counts per module and your reconstructed roster, each person with what they own.
• One-click commit. Once you are signed in, Import into my workspace writes it all in: people seated on your chart, then Rocks, To-Dos, Issues, and Scorecard KPIs (with each weekly value as real history).
• Idempotent. Re-running the import never creates duplicates. It dedupes by title and owner, so you can import again safely after fixing an export.
• Honest about the gaps. The flow tells you what Ninety itself loses on export (Issue comments are dropped; Scorecard history is bounded by the date range you set), and reminds you to export before you cancel Ninety, since Ninety disables export access the moment a subscription lapses.

Where OTP sits next to the frameworks you already run

We also published an eight-part comparison series: OTP vs Scaling Up, OKRs, 4DX, Holacracy, Agile and Scrum, Lean and Six Sigma, V2MOM, and the Great Game of Business. The through-line is the same in every one: OTP is not a rival framework. The framework decides what the company should do; OTP is the operating layer underneath where the work gets executed by a team that is now part human and part AI, each seat with a scorecard and a KPI. Start at the series index.

Two days, fifteen commits, one root pattern. OTP was built first as a tool for a founder running their own org, and that meant the codebase had founder-only assumptions sprinkled across dozens of routes. The moment a second real human signed in -- a teammate joining a live weekly leadership meeting -- those assumptions started leaking the founder's data into the teammate's view. The remediation is a structural tightening: any check that gates data on "who is looking at this?" now reads from the request's resolved member, honors super-admin impersonation, and applies a single canonical role tier from the chart-permissions service.

Multi-tenant scoping across the dashboard

Every route that returns user-scoped data was checked against the rule "must scope by the requesting member's role and chart position." Findings shipped as a coordinated sweep:

• /dashboard: rocks, KPIs, to-dos, delegated lists are scoped to the viewer's claimed chart tiles plus their email. The agent-pushed founder tile is only added to the candidate list when the requester actually IS the legacy founder.
• /dashboard/kpis: scoped to KPIs the viewer owns plus KPIs on teams the viewer is a member of. Owner / admin / implementer keep full-org visibility.
• /dashboard/members: scoped to the viewer plus everyone in their reports-to subtree. Inviters at integrator / visionary / manager rank get a scoped invite picker -- they can still onboard their cone but can no longer see the rest of the org's roster.
• /dashboard/teams: scoped to teams the viewer is a member of via team_memberships. Empty list when the viewer is not on any team.
• /dashboard/team (chart): nodes and edges filtered to the viewer's viewable tile set. Owner / admin / implementer see the entire chart; manager / integrator / visionary see their own seat plus their reports-to subtree; managee sees only their own seat. The same filter was applied to the underlying /api/v1/team/graph endpoint so a client-side d3 fetch cannot bypass the page-level scope.
• /team/review (People Review): identity now resolves through the impersonation-aware path so "view as <user>" rates that user's reports, not yours.
• /team/:externalId profile drill-in (page + JSON API): rocks, to-dos, and tickets for a given chart tile are now gated by canViewTile. Hitting the URL for a tile you cannot see returns a 404 (existence is private).
• "My Practice" nav: the tab now requires an org-wide role (owner / admin / implementer / visionary / integrator). Managers, managees, and members no longer see the practice tab even when the org has a claimed consultant profile.

Impersonation: view-as that actually views as

Super-admin "view as <user>" was leaking the admin's data into the impersonated view because four separate code paths derived "who is the effective viewer?" from the Clerk session (always the admin) instead of from the impersonation cookie payload. All four are now routed through request.impersonation.as || auth.userId so every gate evaluates as the impersonated user. A super-admin-gated diagnostic endpoint at /api/v1/_debug/dashboard-state stays in place -- it returns the resolved identity, role, claimed tiles, and the actual results that would be returned for the current request, so the next tenant-isolation question gets answered in one request instead of guessed at.

The principle going forward

Two memory rules now sit alongside the code so the pattern does not need to be relearned. "Who is the viewer?" gates must read from request.orgMember and request.impersonation.as, never from auth.userId or resolveOrgForUser(auth.userId) (the first two honor impersonation; the second two do not). The canonical "see-all" tier is owner / admin / implementer, defined in services/chart-permissions.ts; Integrator and Visionary roles view through their reports-to subtree, same as Manager. Before any new class of user is invited to OTP -- first paid client, first partner-org, first coach managing multiple orgs -- a structured re-scan runs against the same five greps captured in docs/security-audit-2026-05-27.md. This is a posture, not a one-time cleanup.

Most product bugs surface during demos. This one surfaced inside a real weekly leadership meeting with two people running it for the first time -- which means each fix had a 60-second turnaround window between the bug showing up and being deployed live to the meeting in progress.

The five inside-the-meeting fixes

• Segue / headlines / ratings save no longer errors out. The API auth resolver was only honoring the legacy-founder Clerk path; the invited-member path was missing. Any invited teammate trying to save a field got "Authentication required" even though they could see the meeting page. The resolver now honors both paths, matching what the page-level guard already does. Any invited org member can now run their weekly leadership meeting normally.
• "Add Issue" no longer rejects short titles. The "Add Issue" UI hint promised "any non-empty title" but the server schema required five characters. A four-character title ("Reps" in the live meeting) returned "Add failed: invalid ticket data" with no explanation. Title and description validators were relaxed to one character minimum, matching the UI promise. Spam control already lives in the rate limiter; min-length validation does not need to do that job.
• Saves no longer wipe what the other person is typing. Any save on the meeting page (segue, headline, rating, issue, to-do) fired an SSE event that triggered a full-page reload for every other attendee. The reload preserved scroll position and the focused element's ID, but not the contents of any input the other attendee was typing into. Now every reload path snapshots non-empty textarea / input / contenteditable values to sessionStorage before reloading and restores them on load. Typing in segue while a teammate saves their headline is now safe.
• Meeting to-do due dates accept a custom date. The to-do create form on the meeting page hardcoded the due date to seven days from today with no override field. A new date input next to the priority / recurrence selectors lets you pick any date; if you leave it blank, the seven-day default still applies. Picks anchor to end-of-day local so a "due today" to-do does not show overdue at 9 AM the same day.
• Meeting links in Slack / WhatsApp render as the meeting page, not as a sign-in marketing card. Before today, pasting a meeting URL into Slack rendered the sign-in page's "Free in Beta" mascot card because the unfurl bot followed the auth redirect and scraped the sign-in meta tags. Now the meeting URL returns a public preview at the same path: meeting title, scheduled date in ET, attendee names, and a "Sign in to view" CTA. No agenda contents, scorecard, rocks, ratings, segue, or headlines leak -- public-safe metadata only. Sharing a meeting link with a teammate finally looks like sharing a meeting link.

Personal queue stopped leaking the founder's tasks

The same day, the /me/todos personal queue was found to hardcode the founder's external ID as a fallback owner -- a leftover from when the only user of OTP was the founder. Any invited org member who navigated to /me/todos got the founder's full personal to-do queue grafted onto their own. The fallback now only fires when the requester actually is the legacy founder; everyone else's queue scopes strictly to their own claimed tiles plus their email.

Org chart visibility tightened

The org chart at /dashboard/team previously rendered every node and edge to every authenticated org member. Visibility now matches the existing edit rules: owner / admin / implementer see the full chart, manager sees their own seat plus their reports-to subtree, managee sees their own seat only. The filter applies to both the page render and the underlying /api/v1/team/graph endpoint so a client-side d3 chart cannot bypass the scope by hitting the API directly.

The to-do system is the most-touched surface in OTP, and until today there was a split: the platform had /me/todos, but the AI agents pushed their action items to Todoist, so the queue was actually in two places. As of today, it's one place -- and that place is built for the phone.

The mobile queue is the queue

The /me/todos page was rebuilt as a thumb-reachable mobile app: sticky bottom add bar that's always visible (one input + send, no priority/date upfront -- those live in the per-todo Edit drawer), swipe-left-to-complete on any card, iOS safe-area padding so the FAB clears the home indicator, and a PWA install nudge so you can drop the app on your home screen and treat it like any other phone app. Theme color matches the rest of the brand (Ollie lime). Desktop layout is unchanged.

Your AI agents create todos directly

Every agent on the Sneeze It army -- Radar, Pepper, Dirk, Arin, AIO, and the rest -- has been migrated off Todoist and onto OTP's API. When an agent identifies an action item for you, it lands on your queue in seconds, tagged with the agent's name so you can see which one pushed it. The migration replaces 50+ touchpoints across the agent system; the canonical path now is a single shell wrapper (~/.claude/otp-todo.sh) every agent calls.

Descriptions with formatting

The to-do create form (on /me/todos, /dashboard, and inside weekly leadership meetings) now has a description field with a small toolbar: **bold**, *italic*, and [label](url) link insertion. Cmd/Ctrl+B / +I / +K shortcut too. What you type is what gets stored -- no editor library, no contenteditable surprises -- and the renderer is gated so a malicious link can't inject javascript: URLs.

Todoist is deprecated

Todoist was the bridge while OTP built its own todo layer. That bridge is no longer needed. A migration script (~/.claude/scripts/import-todoist-todos.sh) mirrors any remaining open Todoist tasks into OTP on demand -- run once when you're ready to flip the switch. Agent surfaces still in the codebase reference Todoist only in DEPRECATED sections so the history stays readable.

Four threads shipped today, each as a separate deploy.

Brand pass across every dashboard

The full Ollie palette -- lime green #A8E63A for accents, deep blue #2563EB for action, a warm #F5F7FA page background, and the half-organic / half-circuit mascot -- now runs across all 17 dashboards. The CEO view at /dashboard/ceo got a hero treatment: 96px coach mascot, a green "CEO View" pill, amber attention badge when something needs you, and a real display headline instead of a row of pill chips. Main /dashboard was rebalanced into a Ninety-style 3-and-3 grid (Headlines / Rocks / KPIs on the left, To-Dos / Issues / Agents on the right) so the right column no longer towers over the left. The four coaching panels -- Waiting on others, Founder Dependency, Accountability Gaps, Hand-off candidates -- moved into a single collapsible "OTP Insights" tray below the grid. The My Agents card became a one-line summary that expands on click, opening state remembered across loads. Org-chart tile height bumped from 80 to 104px so employee tiles with two-line roles and contact pills stop clipping.

People Review: only the seats that report to you

The People Analyzer grid at /team/review was showing every human in the org, which made the page noisy and out-of-frame -- you rate your direct and indirect reports, not your peers or yourself. The grid now walks the reports_to subtree from whatever seats you have claimed on the chart and filters to that set, excluding your own tiles. Empty states added: "claim your seat" when you aren't on the chart yet, "no one reports to you yet" when you are. Same People Review was added as a visible banner on the daily dashboard and a "Review People" link in the CEO view's Team section, so it stops being a hidden feature.

Seat Fit and People Review saves no longer 500

The PUT endpoints for seat fit (/api/v1/seats/:externalId/fit) and seat responsibilities were 500-ing because the audit-log insert was sending the seat's varchar external ID (e.g. HUM_DAVIDSTEEL) into audit_logs.entity_id, which is typed as a Postgres uuid. The same bug surfaced through People Review's G/W/C columns as "Could not save rating." Fixed by passing the seat's external ID through details.seatExternalId on the audit row instead, where it belongs. Both save paths now work end-to-end.

Chart claims self-heal on every signup path

Two long-standing bugs in the chart-claim flow closed. First: when a teammate was pre-added to a team before signing up, the system created a synthetic stub member row (clerk_user_id starting with chart:) so the chart could attach to a team membership. When that person later signed up via Clerk, nothing merged the two rows -- the stub and the real account lived in parallel, and the chart had two of them. Second: owner signups went through /onboarding/profile, which created the org_members row with no claim and then called the chart-placement function -- which built the chart entity with the owner's email but never linked it back, so the owner showed up on the chart but their account stayed unclaimed (the exact reason People Review thought you weren't on the chart).

Both are fixed by a single reconciler that runs on every signup path. For any member with an email, it finds chart humans whose contact_email matches, merges any chart stubs pointing at those tiles in (moving team memberships, deleting the stub), and replaces the member's claim with the matched set. Wrong primary claims get overwritten. Real Clerk users with conflicting claims are deliberately left alone -- silent merging of two real accounts is too risky to do automatically. Backfill script scripts/reconcile-chart-claims.ts ran across all existing orgs and caught one silent bug-bite (a member who had been unclaimed for weeks).

The front door has been quietly sharpened over the last three days. The conversion pages are the place new founders first decide whether OTP is real -- they should look like the rest of the product, not like a generic Clerk template glued onto an old layer.

Sign-up: v7 editorial rebrand

The /sign-up page now matches the home page visual language: light editorial backdrop, mono [OTP] lockup in a minimal nav (logo + Sign in, no exit ramps), big Schibsted Grotesk hero, and Clerk's SignUp widget themed to the Ollie green primary on dark ink text. A "What happens next" sidebar with hand-1 / hand-2 / hand-3 numbering tells new founders the three steps after submission. Loading skeleton inside the Clerk slot replaces the empty-box-while-the-widget-loads dead time.

Sign-in: matches the new sign-up, plus a live What's New tile

The /sign-in page got the same minimal nav, "Welcome back" hero with the operating-platform thesis as subhead, and a sidebar tile that auto-pulls the latest entry from this changelog. Returning users see what shipped recently the moment they sign in -- the platform feels like it is moving because it is.

Post-signup flow: layout conflict fixed, all 7 onboarding steps operational

The 7-step post-signup onboarding wizard (profile, team, goals, KPIs, agents, teams, first meeting) had a layout-rendering conflict that was returning an error on the first step for new accounts. Rebuilt on a manual ejs.renderFile pattern that bypasses the global-layout collision; the full wizard now flows end-to-end. Mobile sign-up was verified with a real submit -- Clerk's Cloudflare Turnstile fires correctly, the verification email lands, and the post-signup wizard renders.

Google Ads attribution unblocked

The Content Security Policy was blocking googleads.g.doubleclick.net and www.google.com/ccm/collect, so even successful sign-ups never reached Google Ads as a conversion. Added those origins to the script-src and connect-src directives. The server-side conversion upload that fires on /onboarding/profile (gclid-based, idempotent via the conversion_log table) is now end-to-end operational.

The org chart has been a picture of who reports to whom. As of today it runs the people side of the operating system: who fits their seat, who is overloaded, where accountability is leaking, and how dependent the company still is on the founder. Four updates shipped today, each as its own deploy.

Accountability Gaps -- now actionable

The Accountability Gaps panel on /dashboard used to just list problems. Now it fixes them inline:

• A seat with no measurable gets + Add KPI / + Add Rock forms right in the panel, pre-targeted to that seat.
• Work owned by someone not on the chart gets a Reassign picker that moves it to a real seat.
• An agent with no human above it gets a Set manager picker.
• A filter bar -- All, All Human, All Agents, Direct Reports, No Seat -- narrows the panel to what you own.

Seat Fit

Every seat profile now carries a Seat Fit rating: does the person Understand the role, do they Want it, do they have the Capacity for it. Click to cycle each axis; it saves on the spot. A clean, honest read on whether the right person is in the seat.

People Review

A new grid at /team/review: every person down the side, the three Seat Fit axes plus each of your organization's values across the top. Click any cell to rate it. Each row gets a computed verdict -- solid, needs conversation, or wrong seat. Your value list is managed right on the page.

Delegate-and-Elevate + Founder Dependency

• Hand-off candidates -- the dashboard flags anyone carrying a heavy recurring-to-do load, with a prompt to push the work down to a lower seat or an agent.
• Founder Dependency -- a tile showing the share of all open work (priorities, KPIs, issues, to-dos) that still sits on the top seat. The number that tells you whether the team is actually absorbing the load.

Shared KPIs

A KPI can now be assigned to several people at once. Each person has their own goal; the scorecard sums the goals into one total and sums what each person reports into one actual. On /dashboard/kpis, hit + person on any KPI, pick who and their goal -- a "Shared KPIs" section then shows the rolled-up total with the per-person breakdown.

Today shipped fifteen contained pieces that together turn the Founder 25 cohort from a marketing line into a working product. The thing the platform has been missing -- a real reason for a coach to come back tomorrow -- is now there.

For coaches: the Practice dashboard

• /dashboard/practice -- your coach view across every client org you bring in. Three states: intro (not a coach yet), empty (share-link command center with email + LinkedIn share CTAs), populated (stats + side-by-side client list).
• /dashboard/practice/client/:id -- read-only view of one linked client's workspace. Shows team members, KPIs, AI agents, OOS file count, plus a "Send activation nudge" button that fires a templated email to dormant clients with a concrete first-step CTA.
• "My Practice ★" nav tab -- appears automatically in the dashboard nav for any claimed coach (coach-status detected server-side via a request-level preHandler).
• Send invite by email form on the Practice dashboard -- type a client's address plus optional first name, fire the invite directly from the platform. Reply-To routes to the coach so any reply lands in their inbox.

For coaches: the Founder Certified Coach badge

• Downloadable PNG at three sizes -- full quality (1.3MB) for print/email-signature, web preview (213KB) for site embedding, and compact (66KB) for inline markers.
• Displayed automatically on the coach's public /expert/:slug profile (top-right of the hero on desktop, centered above the headline on mobile).
• Highlighted in the /claim/:slug/done page with two download buttons (full + web).
• Featured on the new /coaches directory and on the /coach Founding 25 perk card.

The /coaches public directory

A new page at orgtp.com/coaches shows every claimed Founder Certified Coach as a card -- name, location, headline, bio, expertise tags, badge. The dark hero shows "X of 25 seats filled" so prospective coaches see real cohort progress, not vapor.

For clients: /welcome quickstart + revocation control

• /welcome -- when a client accepts a coach's invite they now land on a 3-card activation guide ("Add yourself · Add one teammate · Add one AI tool"). Cards flip to a green ✓ as the work gets done. Personalized with the coach's name when attribution is found.
• /settings/coaches -- clients can revoke any coach's access from one page. Founder attribution stays attached for commission accounting; only data visibility is revoked. Two-table design (access revocable, attribution immutable) so commission survives even if the client fires the coach -- same hook GHL agencies use to overcome the "what if they leave?" fear.

The full loop end-to-end

1. Cold email lands → coach hits /claim/:slug → claims their Founder profile.
2. Auto-publish + auto-fill Clerk email → public profile is live immediately, no checkbox to remember.
3. Welcome email fires with the Founder badge, the share link, and a 15-min Calendly CTA. Reply-To routes to David's real inbox.
4. Coach lands on /dashboard/practice → copies their /join/{token} link or fires the invite-by-email form.
5. Client clicks the join link → coach-branded landing → accepts.
6. Two records get written: access (revocable by client) and attribution (perpetual).
7. Client gets a welcome email + lands on /welcome → builds their first 3 seats → coach sees stats populate in their Practice dashboard.

First real signal

Joel Swanson, a business operating system implementer in Colorado, became the first cold-email-driven claim today -- about four hours after his email landed. He wrote back asking the basic "who are you, what is this, why should I care" questions, which became the spec for the new /coach 101 panel and the welcome email template every coach receives from this point on.

Under the hood

• Two new tables -- coach_client_attribution (immutable except by admin transfer) and coach_client_access (revocable by the client). Different mutability rules, intentionally separate.
• Self-healing schema migration on boot via ensure-coach-clients.ts.
• Resend open + click tracking enabled for the founding-25 campaign. Every cold email since this morning gets per-recipient engagement data, filterable by campaign=founding_25_coach.
• Three new branded OG images (1200×630) so /join/{token}, /coaches, and /welcome all show real preview cards when shared on LinkedIn / iMessage / Slack.
• Takedown handler now auto-unpublishes the profile immediately when a coach hits the "remove me" link (the email promise was "one-click removed" but the old code only filed a ticket and waited for a human).

Founder badge for any claimed coach: download the PNG. The Founding 25 still has open seats -- see /coach for the playbook.

Orger.ai shipped today with a new homepage, a mascot named Ollie, and a free builder that lets you drag and drop the humans you have and get grounded recommendations for the AI agents you should hire next.

The bigger news for the network: Orger is the first product built on OTP, not just one that consumes the API.

What that means concretely

• Same Clerk auth as orgtp.com.
• Same Postgres database backing every chart on the network.
• Same MCP server surface, queryable from any MCP-aware AI client.
• Same team graph schema. Seats you create on Orger are OTP seats.
• Same scorecard surface and the same correction loop every other agent on the network learns from.

Why this matters for the framework story

OTP has been called a protocol, a coordination layer, a network, and a transactive memory system. All accurate, none of them prove the framework claim on their own. A framework only earns the word once a product is built on it that could not have been built any other way at the same cost. Orger is that product.

Read the full launch story: Orger is the first site built on OTP.

If you are building something that needs humans, agents, seats, scorecards, and corrections to live on the same protocol, the framework is open. The MCP is npx -y @orgtp/mcp-server. Reach the team if you want to be the second.

The Dashboard shipped earlier today as a full daily manager surface. Tonight is the first polish pass -- the rough edges from a couple hours of real use.

To-Dos: full CRUD inline

• Every to-do row now has Edit, Done, and Delete alongside the checkbox.
• Edit toggles an inline form for title and description -- save reloads, cancel reverts.
• Delete uses a 2-tap confirm (tap once, button turns red "Confirm?", tap again within 3 seconds to soft-delete). No accidental losses.
• Descriptions now display in the row when present, no separate page needed.
• The same UX is live on the mobile /me/todos PWA so the queue feels the same on phone or laptop.

Issues (IDS): status filter, default Open

• New pill bar above the issues list: Open / Identified / Discussed / All open, each with live counts.
• Open is the default selection so the dashboard opens to the issues that still need to be identified first.
• Pure client-side filter -- switching is instant.

MCP-live pill is now clickable

• The amber "not connected" banner already linked to /settings/api. The green "live" pill was static.
• Now both states are clickable surfaces to the same MCP management page. One tap to verify, rotate, or copy the key.

My Agents: compact rows + sourced from the team chart

• Each agent renders as a single line: status dot, name, slim score bar, score, KPI count. ~14 agents drops from ~2000px to ~350px while keeping the at-a-glance "who's red" read.
• Header summary shows total seats, average score, and counts of red/yellow agents.
• Click any row to jump straight to /dashboard/team.
• Bigger fix: the panel now sources from the same team graph that powers /dashboard/team instead of a separate upload table. Score reads from the entity's maturity_level (the Bassim-style maturity score), and the KPI count joins live from your KPI registry. Agents you add to the chart now appear here automatically -- no separate upload step.

OTP's dashboard used to be a publisher splash for owners and a near-empty page for everyone else. It is now a daily manager workspace, designed to be opened every morning.

What you get on /dashboard

• Meeting selector -- pick the meeting you are running (defaults to the next upcoming).
• Headlines -- any attendee can post a headline. The Integrator gets a checkbox to mark each one read; everyone else sees the read state. No more headlines getting lost in a shared text blob.
• My Quarterly Priorities (Rocks) -- this quarter only, filtered to what you own. One click toggles on-track / off-track.
• My KPIs -- the KPIs you own. Type a value, hit tab, it saves to this week's period. The same numbers feed your scorecard.
• My To-Dos -- add inline with a due date, check off when done. Overdue items go red.
• Issues (IDS) -- add an issue right from the dashboard. The Integrator can mark it solved and auto-link it to the current meeting.
• My Agents -- upload a CLAUDE.md or Agent.md file (paste it or pick the file). Each agent gets a 0-8 score: +2 if your org's MCP key is live, +2 if KPIs are declared, up to +3 for recent agent runs, +1 for a description. The score is meant to push you toward connecting MCP -- that is where OTP actually starts working for you.
• MCP banner at the top -- green when an API key on your org has been used in the last 7 days, amber CTA when not. The single most important conversion moment from passive viewer to active user.
• Multi-org dropdown -- if you are a member of more than one org, switch between them right at the top of the page.

Two new roles: Visionary and Integrator

• Added Visionary and Integrator as first-class roles alongside Owner / Admin / Manager / Managee. Visible in the Members invite dropdown.
• The Integrator role unlocks running-the-meeting actions: marking headlines read, overriding a Rock's status, marking issues solved. Owners, Admins, and Implementers also have these powers, so nothing changes for orgs that have not adopted the role yet.

Wayfinding: Dashboard dropdown + tab strip

• The global nav now has a Dashboard ▾ menu (signed-in users) with shortcuts to Daily / Meetings / Team chart / KPIs / Operating plan / Members / Publisher / MCP. One click to anywhere from any page.
• A sticky tab strip renders on every /dashboard/* and /l8/* page so you always see where you are and what other surfaces are available -- role-filtered (managee does not see Publisher).
• The legacy publisher dashboard (OOS files, claims, network learnings) moved to /dashboard/publisher and is one click away from the Daily view.

Why this matters

• OTP works because you use it every day. A login that lands on a publisher splash does not earn a daily habit. A login that lands on your Quarterly Priorities, your KPIs, your headlines, and your agents does.
• The "My Agents" score is a forcing function. Upload your CLAUDE.md, see the score, see exactly which lever (connect MCP, declare a KPI, run the agent, write a real description) moves it. Every visit is a nudge toward making the system live, not decorative.
• The same data that drives your dashboard is the same data your weekly L8 meeting reads. No double-entry, no separate system.

How to start

1. Sign in and visit /dashboard.
2. Pick a meeting from the dropdown. Add a headline. Type a KPI value. Add a To-Do.
3. If the MCP banner is amber, click Set up MCP and create an API key.
4. Upload your CLAUDE.md under "My Agents" and watch the score change as you connect things up.

The weekly leadership meeting is now the L8 Meeting. The name comes from Level 8 of the 8 Levels of Agentic Engineering -- Autonomous Agent Teams. Same agenda shape you already run (Checkin, Scorecard, Quarterly Priorities, Headlines, To-Dos, Issues, Conclude); different reason to run it: every week the meeting forces the question "are we more agentic this week than last week?"

Old /l10 URLs continue to work via 301 redirect, so any link, bookmark, or doc that points at /l10 will land on /l8 automatically. The glossary keeps the old name alongside the new L8 entry so the earlier terminology stays discoverable.

OTP now runs your weekly leadership meeting. Visit /l8 to see your meeting list, or click the new Run L8 Meeting button on your team chart at /dashboard/team.

What you get

• Single-page meeting runner with all 7 sections on one screen: Checkin, Scorecard, Quarterly Priorities, Headlines, To-Dos, Issues, Conclude.
• Per-section timer on a single-line agenda nav. Click any section to jump and start its budget timer. Goes amber under 60s, red and pulsing when over.
• Attendees & Access panel at the top of every meeting. Edit who is invited inline.
• Scorecard grouped by team: Leadership Scorecard (humans), Agent Scorecard (one row per AI agent), and any custom group you push KPIs into. Goal column shows symbols, on-track checkmarks colored green/red.
• Quarterly Priorities with on-track/off-track toggle, owner, status note. Add new ones inline with the new + Add quarterly priority button.
• Issues with full identify-discuss-solve workflow. Solve opens a form that captures the resolution and (optionally) auto-creates a 7-day follow-up to-do for an attendee. Edit, delete, convert-to-todo, and assign ownership all live on each issue. Inline + Add issue button at the bottom of the section.
• Cascading message that builds itself. Every solve appends a one-liner. The Rebuild draft button regenerates the full recap from headlines + solved issues + new to-dos so the team always knows what got decided.

The org chart now shows accountability

• Click any person or agent on /dashboard/team and the edit drawer opens with a Meetings & Accountability block: tiles for quarterly priorities owned, open to-dos, open issues, and meetings attended. Below that, upcoming and recent meetings with status badges and contribution chips.
• Click View full profile for the dedicated person page at /team/:externalId: hero summary, currently-owned items, and a vertical timeline of every meeting they have attended with per-meeting contribution counts (to-dos owned, issues solved).
• This turns the org chart from a static who-reports-to-whom diagram into a real accountability dashboard. Click anyone, see what they own, see what they delivered.

Why this matters

• Most leadership-meeting software is heavy, complicated, and built around someone else's framework. This is built around the work the team is already doing -- Sneeze It is running its own Tuesday leadership meeting from this exact page.
• The meeting layer reuses the same KPIs you push via tally, the same agents on your org chart, the same issues your team raises. No double-entry. No separate system.
• Every AI agent you have can be invited as an attendee, can own a quarterly priority, can own a to-do. Run a real weekly meeting with a mixed human + agent team.

How to start

1. Go to /dashboard/team, click Run Weekly Meeting top right.
2. Create a meeting (defaults to a leadership meeting tomorrow 9am).
3. Add your team as attendees -- humans and agents both.
4. Click Start Meeting. The scorecard and quarterly priorities snapshot at start, so the agenda is frozen at meeting time.
5. Walk through the 7 sections. Solve issues. Cascade decisions. Done.

This is the kind of thing OTP was always meant to be: where your AI team and your human team coordinate from the same surface.

The fifth tool in the public OTP toolbox just went from "promised on the homepage" to "live and scoring." Take it at /checkup.

What it actually does

• 24 questions, 3 per level across all 8 levels of agentic maturity. One question at a time, snappy progress bar, no walls of text.
• Final score is calculated using Bassim's hierarchy rule: a weakness at any lower level caps your score regardless of higher-level capabilities. A perfect L8 with a broken L4 still scores you a 4.
• You get a single number out of 8.0, a tier headline (Tourist, Tinkerer, Operator, Orchestrator, Autonomous Agent Team), per-level breakdown showing which level capped you, and a personalized roadmap with the three highest-leverage moves for your next jump.
• Result is also emailed so you can share it with your team or come back to it later.

Why this exists

• Most teams cannot answer "where are we, really?" on AI maturity. Vendors sell against the top of the ladder. The honest read is harder to come by.
• The same scoring engine our internal evaluator (Bassim) uses to score Sneeze It nightly is now public, translated from "reads your agent files" into "you tell us what is true."
• The roadmap is the point. The score is just the way in. If you walk away with three concrete moves to make this quarter, the quiz worked.

The 8 levels, briefly

• L1 Tab Complete -- AI as autocomplete in your editor or inbox.
• L2 Agent IDE -- you have built or deployed at least one real agent.
• L3 Context Engineering -- system prompts, rules files, persistent context.
• L4 Compounding Engineering -- lessons learned compound; agents read your playbooks.
• L5 MCP & Skills -- AI is wired to real tools and takes real actions.
• L6 Harness Engineering -- validation, observability, authority bounds.
• L7 Background Agents -- agents run on a schedule without you starting them.
• L8 Autonomous Agent Teams -- agents talk to each other; you are not the message bus.

The homepage and /tools CTAs both repoint to the live calculator. The Checkup is free, no signup required to take it -- only to see the result.

Three quiet improvements to the public site, all shipped together.

A new /start-here page

• Modeled on the classic one-promise, three-steps, one-calendar pattern.
• Hero promises a 30-minute conversation with David Steel, founder of OTP. No slides, no sales gauntlet.
• "What happens on the call" section walks through the three things in order: map your AI footprint, find the coordination gaps, decide the next move.
• Inline Calendly widget renders the full calendar plus time slots without scrolling, branded against the OTP color palette.
• Three reassurance statements address the obvious objections: it is free, you are talking to the founder, we will tell you if OTP is wrong for your situation.
• A new "Schedule a 30-min intro" primary button is now in the hero of /why-otp, /what-is-otp, and /tools, all routing to /start-here.

Heroes that put the CTA above the fold

• The hero on /, /why-otp, /what-is-otp, and /tools all use the same 12-column grid now: text on the left (col-span-7), an editorial illustration on the right (col-span-5).
• H1 sizing tightened so the CTAs land in the first viewport without scrolling on a standard laptop.
• Each illustration is a hand-drawn line-art piece that maps to the page's specific copy: an org chart with disconnected AI tools (why), a four-layer Model/Protocol/Network/SaaS stack (what), an open toolbox with five tools (tools), a calendar slot and 30-minute clock (start-here).
• All illustrations are 1024px lossless WebP with transparent backgrounds. The page background shows through cleanly.

Accessibility and craft pass

• Site-wide prefers-reduced-motion gate so animations honor user preference. Vestibular-disorder users and low-end devices get a calm site.
• Keyboard users can now tab into the Explore and Learn nav dropdowns; previously hover-only.
• Email signup field has a visually-hidden label and an aria-live result announcement for screenreaders.
• theme-color meta added so mobile browser chrome matches the dark nav instead of defaulting to system white.
• transition: all antipatterns replaced with explicit property lists across the layout.

None of this changed the product. It changed how the product introduces itself.

Adding humans one tile at a time was fine for a 10-person team and painful for a 50-person team. Today the chart gets a CSV import path that scales.

Two templates, your choice of detail

• Simple: name, role, reports_to. Three columns, enough to draw the org chart and nothing else.
• Full: name, role, contact email, contact phone, Slack ID, reports_to, job description, authority level, skills, MCPs, status. Every field a human tile carries on the chart, mapped one-to-one.
• Both templates download as real CSVs with three example rows. Open in Sheets / Excel / Numbers, edit, save, drop back.

Reports-to resolves the way humans actually write it

• Match by name first against existing humans on the chart and against rows in the same import. So a row "John Smith reports_to Jane Doe" works whether Jane is already on the chart or being created in the same CSV.
• Falls back to external ID (e.g. HUM_JANEDOE) for power users who already know them.
• Names that match nothing leave the parent unset and surface a warning in the result summary -- the row still imports, you just see exactly which links did not land.

Two import modes, one of them destructive

• Addition (safe): upsert by name. Existing humans whose names match get their fields refreshed. Humans not in the CSV are left alone. Default mode, never deletes anything.
• NEW / Overwrite: upsert by name, then delete any human not in the CSV. Agents that escalated to a deleted human lose that link automatically; other humans that reported to a deleted human do too. Confirmation prompt before any destructive run.

Preview before you commit

• The drawer parses your CSV client-side and shows a row-by-row preview: which rows are new, which are updates, and which references will resolve.
• A live count tells you exactly how many humans NEW mode would delete -- before you run it.
• The whole import lands as a single atomic write to your draft OOS, not 50 round-trips. Up to 500 rows per file.

Find it next to "Invite member" at the top of /dashboard/team. Owners only.

Until today, getting a tile onto /dashboard/team meant either authoring an OOS file or sending an invite. That worked when you had real teammates to email, but slowed you down when you wanted to model an agent or place a known human on the chart without bringing them in yet. Two changes fix that.

Two new entry points to the create flow

• "Add to chart" button at the top of the chart, next to "Invite member." Opens the same side drawer you already use for tile editing, but in create mode -- empty fields, a Human / Agent toggle at the top, and a Create button instead of Save.
• Per-tile "+" button appears in the bottom-right corner when you hover any tile (owners only). Click it and the create drawer opens with the tile you hovered preselected as the parent -- "reports to" for humans, "escalates to" for agents. One click adds a direct report under the seat you are standing on.

Same drawer, same fields, same SOP section

• Every field from the edit drawer is available in create mode: role, mission, job description, authority level, agentic maturity, platform, status, contact email/phone, Slack ID, skills, MCPs, SOPs, KPIs.
• The Invite section and Delete button are hidden in create mode -- the rest of the panel is identical, so the visual rhythm of the chart is unchanged.
• Submit creates the tile in your draft and saves all the extra fields in one go. Reload to see it on the chart.

Agent Builder runs inside the chart now

• In agent create mode, a "Use Agent Builder" button appears at the top of the drawer. Click it and the Agent Builder slides in from the right -- same rail as the create drawer, white background, light theme.
• Just the wizard. No hero, no marketing sections, no footer. Industry, title, description, skills, tools, personality framework, review.
• On generate, the result populates the create drawer underneath: name, role, mission, skills, MCPs all filled in, with the full generated CLAUDE.md added as a SOP entry titled "Generated CLAUDE.md." The Agent Builder closes; you adjust anything you want and click Create.
• The standalone /agent-builder page is unchanged for visitors who land on it directly.

Why this matters

The seat is the unit of an org chart, not the body. Until you can put a seat down without sending an email, the chart is gated by your willingness to bother people. Now you can model the team you want -- humans you have not invited yet, agents you are still drafting -- and the chart is a working spec from the moment you put a tile on it. The Agent Builder integration closes the loop between "I need an agent for X" and "the agent is on my chart with the right SOPs inherited from its parent." Two minutes from idea to placeable seat.

Today is the pivot. OTP started as the Organization Transport Protocol for AI agents. The acronym was right; the scope was too narrow. The agent army is a slice of the organization. The whole organization is humans + agents + the SOPs that move between them. That is what OTP is for now.

SOPs as the unit of coordination

• Author once: click any agent or human on /dashboard/team and add SOPs in the side panel. Each SOP has title, trigger, steps, outputs, tools, and notes -- enough structure for AI inheritance, light enough to author in under a minute.
• Five Founder/CEO templates seeded: daily inbox triage, weekly L8, monthly stakeholder update, founder-led discovery call, quarterly Rocks-setting. Click "+ From template..." in the SOP section, pick one, edit to your voice, save. Sixty seconds to a working operating cadence.
• Purple "N SOPs" badge on every chart tile that has authored SOPs. Visual confirmation that a tile carries executable spec, not just a name.

AI agents inherit SOPs from their human

• Inheritance: when an agent escalates_to a human, the agent inherits that human's SOPs as runtime context. No copy-paste, no re-explanation when a new instance spins up.
• Copy as Agents.md / Claude.md: every agent's edit panel has a one-click button that compiles own SOPs + inherited SOPs + role/mission/authority into a markdown file, copied straight to your clipboard. Pick the format your stack expects -- AGENTS.md (cross-platform default), CLAUDE.md (Claude Code), .cursorrules (Cursor), or generic system prompt. Drop it into your runtime and the agent runs on the org's latest accountability state.
• Tooltip: hover any agent and see "Inherits N SOPs from {parent}." The SOP layer is visible at a glance.

Multi-user invitations

• Invite to claim a tile: open any human node and you will find an "Invite to claim this tile" section. Email goes out from notifications@mail.orgtp.com with a 30-day-TTL signed link.
• Invite someone new: the chart header has a "+ Invite new member" button that creates a fresh tile and fires the invite in one action -- name, email, optional role, optional reports-to.
• Accept-invite landing: the recipient lands on a page showing the org name, the tile reserved for them, and the expiry date before they sign in. Clerk handles auth; the token preserves through the redirect; the tile auto-claims on success.
• Pending invites drawer: owners get a header button (with badge count) that opens a drawer of every pending invite, with revoke buttons.
• "Claimed" pill shows on every human tile that is bound to a real member account.

Editor polish

• Contact fields on humans: email, phone, Slack ID, with small contact pills on the chart tile when populated.
• Status field with active / paused / inactive / retired / terminated. Inactive nodes render grayscale with a strike-through name -- a clean way to mark old data without losing structure.
• Explicit Delete button in the edit panel. Removing a tile also scrubs any escalates_to / reports_to / override_authority references that pointed at it, so the chart never dangles.
• Sticky filter: the All / Agents / Humans toggle persists in localStorage. Refresh the page; the view stays where you left it.

What the framing change means

OTP is no longer "the AI coordination protocol." It is the substrate for hybrid human-and-AI organizations. The chart is the surface, the SOPs are the substance, the network is the leverage. /about and the FAQ are updated.

Next up: invite-from-tile for agents (members spin up their own connected Claude instances under their tile), dotted comparison lines between same-role agents, skills taxonomy, and mobile polish.

Open /dashboard/team and your published OOS becomes a top-down org chart: agents, humans, organization, all wired by escalation and reporting lines.

• Live derivation from your OOS: the chart reads entities.agents and entities.humans from your latest draft (preferred) or published file. No new schema, no separate database. Your OOS is still the source of truth.
• Click any node to edit: a side panel opens with name, role, mission or job description, authority level, platform and status (agents), skills, and the escalates_to or reports_to dropdown. Saving creates a draft if you do not already have one.
• Drag and drop to restructure: drag any agent or human onto another box and it becomes the new child. Cycle prevention rejects drops that would create loops. The PATCH writes back to your OOS draft and the chart re-renders.
• Type filter: the All / Agents / Humans toggle re-lays out the tree to show just what you care about.
• Status banner: shows whether you are looking at a draft (with version number) or your published file. Edits never touch your published file until you republish.

If your OOS does not have entities.agents or entities.humans populated yet, the chart will look empty. Reach out and we can help you migrate. Next up: skills taxonomy, mobile polish, and add/delete nodes.

OTP's biggest update yet. We flipped the entry point: instead of asking you to publish, we're giving you something useful first. Industry coordination playbooks written from real production experience.

• 6 Industry Playbooks: Agency (34 practices), Fitness/Franchise (29), Healthcare (24), SaaS (23), Professional Services (50), and E-Commerce (43). Each practice includes what to do, why it matters, and what goes wrong without it. These are original content from teams running 10+ AI agents in production, not scraped or AI-classified consulting reports.
• Download as CLAUDE.md: One click exports any industry playbook as a CLAUDE.md file you can drop into your project. Your agents start following proven coordination patterns immediately.
• CLAUDE.md Generator: Tell us your industry, team size, and agent count. Our AI generates a complete, customized CLAUDE.md using your industry's coordination practices as context. Fallback templates ensure you always get something useful, even if the AI is having a bad day.
• Practice Voting: Upvote and downvote practices on any industry page. Community validation surfaces the practices that actually work in the real world.
• Newsletter: Weekly coordination intelligence updates delivered to your inbox. No account required, just your email. Stay informed about new practices, industry playbooks, and what's working for other AI teams.
• Homepage Refresh: Three-slide hero showcasing industry playbooks, the CLAUDE.md scanner, and newsletter signup. All 6 industries with live practice counts.

The idea is simple: come to OTP, get your industry's playbook, use it. When you're ready to share what you've learned, publish your OOS. But start with value, not a request.

Three new features that make uploading your CLAUDE.md safer, more actionable, and more fun.

• Vulnerability Scanner: Before you can publish, OTP now scans your entire system prompt for sensitive data -- API keys, passwords, database URLs, credit card numbers, SSNs, revenue figures, salary data, employee names, Slack IDs, internal file paths, billing rates, and more. 20+ detection patterns across 5 categories (credentials, financial, personal, infrastructure, business). Critical and high-severity findings block publishing until resolved. Your secrets stay secret.
• Foundation Score (0-100): Instant structural triage on upload. 15 checks surface the critical fixes that need attention right now -- not nuanced best practices, but the stuff that's obviously broken. No human override authority? No escalation paths? No error handling? Agents with no clear responsibilities? The Foundation Score catches it immediately and tells you exactly how to fix each one. Critical failures cap your score regardless of everything else.
• Share Your Score: New share button on both the scan results page and the dashboard. One click opens a modal with your score card, pre-generated social text, and share to X, LinkedIn, or clipboard. Designed for virality -- "Just scanned my AI agent system. Coordination Score: 82/100. What's yours?" Every share is an ad for OTP.

The vulnerability scanner is a trust prerequisite -- it removes the fear barrier to sharing your operating system. The Foundation Score gives instant value on first upload. And the share button turns every scan into a growth opportunity.

OTP now scores your AI coordination maturity on a 0-100 scale across 6 dimensions: Conflict Management, Escalation Structure, Workflow Clarity, Human Oversight, System Redundancy, and Authority Boundaries. The score is the product.

• Scan Results Page: Animated score ring (Lighthouse-style), 6-dimension breakdown bars with color coding, and insight cards grouped by severity. This is the "aha moment" -- upload your CLAUDE.md and see what the scanner finds.
• Fix Issues Inline: Each critical/warning insight has a "Resolve This" button. Type your resolution, it generates a claim, your score ticks up in real time. The scan page IS the diagnostic session.
• One-Click Publish: After fixing issues, "Publish to OTP Network" does everything in one click -- auto-fix, create OOS, publish, redirect to your live page. No confusing intermediate steps.
• Dashboard Overhaul: Your dashboard now shows the same score ring and dimension bars. Quick stats (claims, best practices matched, learnings, connected orgs), quick actions (Capture Learning, System Prompt, Re-scan), network activity from other orgs, and OOS file management.
• Import-First Publish: The publish page now leads with "Upload File" (drag-drop zone for CLAUDE.md) instead of hiding it as a "power user" option. The wizard is secondary for people starting from scratch.
• CLAUDE.md Parser: New /scanner/from-text endpoint extracts agents, systems, workflows, and oversight patterns from raw CLAUDE.md files. No more "Scanner could not analyze this content" errors.
• Copy as System Prompt: One button on any OOS detail page formats your claims into a compact, token-efficient block for pasting into CLAUDE.md. Shows Token Efficiency Index.

The creation experience is now the product. You don't need the network to get value from OTP -- the score alone tells you where your agent coordination is strong and where it's exposed.

OTP is now a living system. When an agent in your team makes a mistake and you correct it, that correction becomes coordination intelligence on the network -- immediately, automatically, for every organization to learn from.

• Auto-capture: When you correct an agent's output, the agent automatically calls capture_learning with what failed, what to do instead, and why. Every correction is a learning. No manual step.
• Auto-publish: Learnings go directly to your published OOS and the OTP network. No draft step, no delay. Your correction is available to the entire network within seconds. (Superseded: publishing now goes through your draft OOS with human review.)
• Auto-pull: Before executing their main task, agents check OTP for relevant learnings: "Has anyone (including other orgs) learned something about what I'm about to do?" If yes, they follow the learning instead of repeating the mistake.
• Cross-org learning: When another organization discovers a coordination failure that matches your agent setup, OTP surfaces it. Their failure becomes your prevention.

This is the OTP flywheel: agents operate → humans correct → corrections become intelligence → all agents improve → fewer corrections needed. Sneeze It is the first organization running this loop live.

OTP is now a multi-publisher best practices platform. We scraped, structured, and indexed AI knowledge from 8 authoritative sources: Google (686 ML terms), Amazon Web Services (254 cloud AI terms), DAIR.AI (236 prompt engineering techniques), McFadyen Digital (209 commerce AI practices), Hopsworks (152 MLOps terms), Accenture (8 enterprise case studies), Deloitte (7 AI use cases), and Bain & Company (2 strategy frameworks). Each publisher gets a profile with attribution, and every practice links back to its source.

• Publisher profiles: A new expert type. Each publisher has a full profile, linked best practices, and attribution on every card.
• OOS-connected matching: Best practices are matched to your OOS using Jaccard similarity with concept synonyms. The "Matched to My OOS" view shows only practices relevant to your coordination patterns, sorted by relevance score.
• Implement and Ingest: Click "Implement" on any best practice to see the full definition, publisher credit, and which organizations align with that pattern. Click "Ingest into my OOS" to add it as a claim to your draft OOS.
• Best practices on OOS detail: Each published OOS now shows a "Relevant Best Practices" section with matched practices, scores, and source links.
• Bidirectional matching: When viewing a best practice, see which organizations implement that pattern -- creating a two-way discovery network between publishers and practitioners.

This introduces a new expert type: Publisher. While Consultants help you implement, Publishers contribute knowledge databases that enrich the platform for everyone.

• OOS limits raised 3x: Maximum word count increased from 5,000 to 15,000 words. Section claim caps tripled (e.g. 10 to 30). Upload limit raised to 2MB. Minimum lowered to 500 words for starter OOS files.
• MCP server auto-detection: The auto-fixer now scans OOS content against 30+ known tools (Slack, Gmail, Google Ads, Meta Ads, Todoist, etc.) and populates the Infrastructure Graph automatically. All published OOS files have been enriched.
• Publish wizard: tool collection: Step 2 now includes checkboxes for 20 common MCP servers/integrations alongside AI platforms. Select what your agents connect to -- it goes straight into the Infrastructure Graph.
• Agentic Level Calculator v2: Expanded from rigid exact-match keywords to 100+ semantic patterns. Checks frontmatter metadata (platforms, MCP servers). All orgs rescored -- most jumped from L2 to L5-L8.
• Sign In page: Dedicated /sign-in page with Clerk integration. Sign In button now visible in the navigation for all visitors.