BookYourPTO

Changelog

Stay up to date with the latest changes, improvements, and bug fixes in BookYourPTO.

BookYourPTO v2.0.0 release notes

Latest

v2.0.0

BookYourPTO Intelligence (Beta)

  • An assistant built into BookYourPTO. Ask about leave, expenses, schedules, documents and your team in plain words, from the Assistant link under Home in the sidebar, the panel at the top right of every page, or the full-page view at /assistant. It answers from your organization's own records and can show the answer as a table, a chart, or a CSV or PDF file to download.
  • It does the work, after you approve it. It can book leave, add an expense from a receipt you attach, log time, schedule a week of shifts, approve or reject requests you name, and more. Every change shows a card to confirm first. You can choose how much it may do on its own: Ask first (the default, every change asks), Accept edits (simple additions run, while approvals, sending and deleting still ask), Auto (everything except high-risk changes) or Plan (it only lays out the steps). Deleting, bulk changes and other high-risk actions always ask.
  • Your own permissions, nothing more. It acts as the person using it, so it can see and change only what that person could in the app. Employees see their own records; managers, department heads and administrators see what their role allows.
  • Questions on every plan, actions on Pro, Business and Enterprise. Free includes questions; paid plans can also make changes. Each person has a weekly usage allowance that resets every week, shared across shorter sessions, and higher plans include more.
  • Attachments, skills and memory. Attach images, PDFs, text or CSV files (up to 5 files, 5MB each). Ready-made skills guide common jobs such as turning a receipt into an expense, planning time off, team coverage and approvals triage. It remembers short facts about how you work only when you ask it to, and you can see and delete them under /memory.
  • Chats you can come back to. Chats get short titles, are listed under Recent, and can be searched by what was said. Type / for commands such as /new, /skills, /usage and /export.
  • In your control. Executives and administrators can switch it on or off for the organization, for departments or for individual people in Settings > Intelligence, and see the company's usage there. Each person sees their own usage on the Usage page. It only works with BookYourPTO: questions about anything else are politely refused, and likely misuse is recorded for administrators under Security violations. Conversations are never used to train AI models.
  • Approval cards say exactly what will change. A card for a change to an expense claim names the claim (report number, title, status and total) and, for a single purchase, per diem or mileage row, that row (description, date and amount). Projects and clients are named too. The details are read with your own access, so a card never shows a record you could not open.
  • Changes that cannot be undone always ask, even in Auto. Auto and Accept edits run most changes straight away; deleting, removing, deactivating and other changes that cannot be undone still show the card, and the card now says why.
  • The page behind the assistant updates itself. After the assistant changes something, the page you have open (Home, Expenses, a calendar…) loads its data again on its own, keeping your place, with a short "The page has been updated" note. If you have typed into the page or have a dialog open there, it is left alone and the assistant offers the refresh instead, so nothing you were doing is lost.
  • Documents look like yours. PDFs the assistant makes contain only what it wrote: no product name, no "made with" line, and no organization name or date added above the title (they appeared twice when the document already had them). The footer shows page numbers only, and the file's author is your organization.
  • Word documents. Ask for a Word or editable document and the assistant makes a real .docx (headings, lists, tables, signature blocks) that opens in Word, Google Docs, Pages and LibreOffice, kept on your Files page like PDFs. A conversation, or a single answer, can also be downloaded as Word as well as Markdown (the download menu, or /export word). Word files carry no product stamp either: page numbers only, with your organization as the author.
  • Documents and answers get names and totals right. A document about you now names you in its Employee or Name field (it sometimes used the organization's name), and counts and totals come from the records themselves, the same in the answer and in the file made from it (a summary once said 42 requests and its Word copy 43), with no rows invented for records that have no dates.
  • Send your next question while an answer is still being written. Questions you send mid-answer used to disappear. They now wait above the message box ("Queued — sends when this answer finishes") and are sent one after another, in order, as each answer ends, with any files you attached. Select one to edit it again, or remove it with the X; up to 5 can wait. The queue waits for you behind an approval card, a pause at your usage limit or an answer that did not finish, offering Send now on each question, and carries on by itself once the approval is answered. Queued questions belong to their chat and are kept only in the open page.
  • The usage-limit notice keeps itself current. After a plan upgrade, a reset, or returning to the tab, the assistant re-checks your allowance and clears "You've reached this session's limit" without a page reload (it used to stay until the page was refreshed).
  • Help improve Intelligence.
    • A thumbs-down can say what went wrong: Wrong answer, Wrong record, Didn't do it, Too slow or Other.
    • At moments that matter (a thumbs-down, an action that failed, a cancelled approval, or a question asked again), the assistant may ask whether you would like to share the whole conversation to help improve it. It asks at most once a week, never twice in a conversation, and "Don't ask again" stops it.
    • Settings > Intelligence > Help improve Intelligence controls this for the organization. It is on by default and only an Executive can change it (administrators see it read-only). With it on, a copy of conversations with names and personal details removed is kept to improve Intelligence, and a full conversation only when someone chooses to share it. With it off, nobody is asked and no conversation text is kept for this purpose.
    • Conversations are never used to train an AI model, never sold, and never shared with other organizations. See the updated Privacy Policy.

Bug fixes

  • Shifts accepted a department, project or task from outside the organization. Creating shifts, one at a time or in bulk, now checks each one.
  • Assigning training or choosing goal-setting people no longer skips names silently. Naming someone who is not an active member used to be ignored (and could empty the goal-setting list); the request is now refused with a clear message. A bulk training due date is now stored as the calendar day chosen.
  • Receipt scanning on the backup AI provider works again. It asked for a model the provider had retired, so every scan sent there failed (and the scan failed outright when the main provider was unavailable).
  • The assistant's mode menu is no longer cut off. On the assistant home page it opened upward under the page header and hid "Ask before acting"; it now opens wherever there is room and never grows past the window.
  • Names only link to a profile you can open. With Calendar View visible, opening a colleague's calendar showed their name as a link, and clicking it ended on a "Forbidden" page. The same happened in a profile's Direct Reports section, on the home page's My Team, Probation Reviews and Next 1:1 cards, and on the documents "View profile" button. Each now links only when you are allowed to open that profile (your own, anyone in your scope as a department head, or anyone as an administrator or executive). Otherwise the name is shown without a link. Next 1:1 opens your own 1:1s when you cannot open the other person's profile.
  • Performance review cycle reminders are sent again. Since review cycles gained reminders, the daily job that sends them failed on every run, so department heads, administrators and executives were never told a review cycle was opening or an employee's review anniversary was near. Reminders now go out, in the app and by email, on the milestones each cycle sets (14 and 3 days ahead by default), once per milestone.
  • The AI connector (MCP server) works again across the product. Many assistant actions had stopped matching the app and either failed or did nothing while reporting success. Signing in through the assistant failed straight after a correct password. Approving or rejecting time entries, adding expense line items, per diems and mileage, assigning or sending documents for signature, bulk sending, offboarding an employee, adding dependents, equity grants and commissions, creating project issues and calendar feeds were all refused. Many lists answered "0 found" when there were results. Every assistant action has been checked against the app and corrected, and each one now describes who may use it and why it may be refused.
  • Updating a Slack or Teams integration through the assistant no longer removes its webhook. Changing only which notifications are sent used to clear the saved webhook address.
  • Billing actions through the assistant. "Cancel immediately" cancelled at the end of the period instead, and choosing yearly billing at checkout billed monthly. Both now do what was asked.
  • Connector sessions no longer break when their sign-in expires. An expired sign-in was answered with a server error instead of "sign in again", so the AI client did not refresh it and the connector stopped working until it was reconnected. It now refreshes automatically.
  • Opting out of (or back into) data sharing no longer shows an error, and the choice is recorded.
  • Undoing a data import no longer reports "Rollback failed" after removing the records.
  • Profiles show work schedule history, which was always empty.
  • Historical rows in custom profile tables can be saved again, on the day chosen.
  • Creating a certification type or asset category with an existing name explains that, or restores the archived one, and archived ones are no longer offered for new records.
  • Saving only the notification types for Slack or Teams no longer disconnects the webhook.
  • Deleting a department no longer leaves its teams pointing at it.
  • Time entries inside a finalized or signed timesheet can no longer be approved or rejected, in the app or through the public API, so a signed timesheet always matches its entries. Any still pending there leave the time approvals queue and every pending count (sidebar badge, Home card, mobile app) as soon as the timesheet is finalized: the signed timesheet is the record.
  • Excel timesheet exports are recorded in the audit log, no longer fail when a name contains characters such as "/" or ":" or two people share a name, and open formula-like text as plain text. Holidays and leave always land on the right calendar day, an export with no dates covers the current month in the organization's timezone, and a range over a year is refused with a message. Names beginning with "=" or "+" appear exactly as typed.
  • A time entry's task must belong to its project; changing the project clears a task from the old one.
  • Bulk-deleting time entries respects finalized timesheets.
  • Clearing a schedule week uses the organization's timezone for which shifts it removes.
  • "Today", "this week", "this month" and "this year" in Time Tracking follow the person's timezone, and month and year totals include their last day.
  • Due dates, signature deadlines, expiry dates and training due dates are saved on the day chosen, in every timezone.
  • "Some employees" onboarding tasks are given only to the people chosen, and automatic onboarding falls back to the default assignee when the template's assignee has left.
  • Department heads can open the task list of an onboarding they manage.
  • A meeting can't be added to a paused 1:1, or to one with a deactivated participant.
  • Importing straight from a connected platform is hidden until it is finished: synced rows never reached the preview. Export a file from the platform and upload it instead.
  • Dates in custom profile tabs no longer show the previous day for viewers west of UTC.
  • Organizations at UTC+12 and beyond: employees see the first day of a published schedule week.
  • The schedule week's last day no longer slips across a daylight-saving change when emailing or copying a week.
  • Two-factor settings no longer offer Disable when the organization requires two-factor, and say why. Department heads see Rename, Move and Delete only on documents they sent, and 1:1 edit and delete controls appear only for people who can manage that 1:1.
  • Manual and edited time entries are saved at the time you enter, in the employee's own timezone. Entries added or edited on the timesheet were stored hours off in organizations outside UTC, and each save shifted them again; saving without changing the times now leaves them alone. The API refuses a time without a timezone (INVALID_INSTANT).
  • Booking a morning or afternoon off for a single day counts as half a day, matching the calendar. It was charged as a full day unless both halves were chosen; for a one-day request the ending now follows the start.
  • Project, task and issue dates no longer show a day early.
  • Clients can be deleted from the client page; a client with active projects asks you to archive or complete them first.
  • Travel (mileage) entries can be edited on an expense claim while it is editable.
  • Time Tracking updates right after you clock in or out (This Week, This Month, the activity log), without a reload.
  • 1:1 notifications and reminder emails show the correct meeting date.
  • "View Audit Trail" works on filed documents (those without a signature request); it showed an error. Signing certificates number signers from #1, the audit trail no longer says "created the document" twice, and the signing screen's two progress counters agree.
  • Report PDF exports show the logo again in production.
  • Export previews offer one honestly named "Download CSV (opens in Excel)" button.
  • Success messages appear after adding, editing or deleting leave types, departments and holiday locations, and after sending, assigning or declining a signature request; a failed send of signing emails says so.
  • New organizations no longer see a "not found" error on their first Home load, and the unlock dialog for a locked date says who the lock covers; the expenses status chart adds up.
  • Every delete and other irreversible action asks in the app's own dialog, naming what will be lost, instead of the browser's pop-up. Deleting a shift from the Schedules list view now works (it did nothing), and deleting from a list view asks once instead of twice.
  • Export menus say what they produce: "Export CSV (opens in Excel)" opens a preview, "Download CSV now" downloads at once, and the security log export no longer saves a CSV under an .xlsx name. CSV exports of users, expense claims and security logs keep formula-like values as plain text.
  • 1:1 reminders go out on the right day in every organization's timezone.
  • Adding a holiday location loads the organization's current year, and past holidays follow the organization's calendar.
  • The "Time entries require manager approval" note shows only to people whose entries need approval. Editing a time entry shows whose it is instead of a picker that could not change it.
  • Smaller fixes: meal allowance tiles show cents; the calendar's year arrows have names and are no longer cut off on phones; Home, Notifications, Support, Usage and Calendar have tab titles.
  • Mobile: manual time entries use your own timezone (or your organization's), so hours entered on a phone set to another zone land on the right day and time; booking a morning or afternoon off for one day counts as half a day.
  • Invalid filters, dates, page numbers and timezones across leave, reports, time, schedules, security logs, notifications, documents, training, 1:1s and profile edits now get a clear message instead of a server error.
  • Add and Edit Time Entry showed two forms at once (from an earlier change in this release). In Single day mode and when editing an entry, the Multiple days fields appeared under Clock In and Clock Out; for a person with their own timezone, Multiple days showed no fields at all. Entries were saved correctly; only the form was wrong.
  • The Expenses list showed totals in the Pre-tax column. Tax is part of a purchase's total, so adding the parts again counted it twice; the column, the page totals, sorting and "Download CSV now" now show the amount before tax, and every amount in that CSV is rounded to the cent.
  • "Resend email" on the verification page now says why it failed. When an email cannot be sent to the address, the page shows that message instead of a generic "Please try again" (which could not succeed anyway).

Security

  • HR records could be added to a person in another organization. Adding a compensation entry, bonus, equity grant, commission, education entry, language or visa to an employee checked only that the person doing it was an administrator, not that the employee belonged to their organization. All seven now refuse anyone outside the organization, and a guard test keeps every "add to a person" action checking it.
  • Education and languages are now private HR records. Any employee could read any colleague's education history and languages. They are now visible to the person, administrators and executives, and department heads for their departments, like certifications.
  • Emailing a schedule could reach the whole organization. Choosing a department that matched no one sent the schedule to every scheduled employee, and a department head could email any department. An unmatched department now sends nothing, and department heads can email only their own departments.

A whole-app review of every server endpoint, looking for the same kinds of mistake, found and fixed the following.

Organization boundaries

  • Notes can only be added about people in your own organization. An administrator could add a note about a person in another organization, and that person was emailed about it.
  • QuickBooks employee, vendor and department links accept only people and departments in your own organization; the mapping screens could otherwise show another organization's names.
  • A person's working schedule can be looked up only within your organization, and by department heads only for their departments.
  • Only the person who opened a support request, or their organization's administrators, can attach files to it.
  • Training courses, certifications and locked dates accept only categories, types, people and departments from your own organization.
  • Another organization's records are answered "not found". An expense claim, line item, per diem, mileage entry, receipt, leave decision, cancellation review or user edit naming another organization's record answered 403, which showed the identifier was real elsewhere. It now answers 404, exactly like an identifier that does not exist; a guard test keeps it that way.

Sign-in and sessions

  • Organizations that require single sign-on now refuse password sign-in on the server for the web sign-in page and connected AI clients, not only by hiding the form. The mobile app keeps signing in with email and password, so people can still set or reset a password for it. Nothing changes until SSO is fully set up.
  • SSO sign-in requires an email the identity provider has verified, and an allowed email domain. With SSO on and no domain configured, SSO sign-in is refused instead of accepting any domain.
  • Only executives can add, verify or remove SSO email domains. The Auth0 domain must be a public hostname, and the default department for new SSO users must belong to your organization.
  • A sign-in session can no longer renew itself indefinitely. Re-checking an already verified email returned a fresh session, which survived signing out, password changes and role changes. Verification links now work once and no longer sign you in.
  • If a sign-in session is used from two places at once, which can mean it was stolen, the person is signed out on every device. Two tabs refreshing together are still treated as normal.
  • Connected AI clients: a connector's sign-in can only be renewed by the client it was issued to, and reusing an old, already-replaced connector sign-in signs that person out everywhere, the same protection the web and mobile apps have. Renewing through the connector can no longer turn a short browser session into a longer one, and a brief cache outage no longer stops connectors renewing.
  • Single sign-on fails cleanly with a clear message when the identity provider answers with a redirect or an error, instead of an unexpected error.
  • Email verification can no longer be skipped. Signing up with an address that cannot receive mail (a made-up domain, for example) used to mark the account verified when the verification email failed to send. A failed send now leaves the account unverified, on the verify screen with Resend and "try a different email", and Resend reports a failed send instead of claiming it went out. Only an installation with no email settings at all skips verification.
  • Two-factor authentication can't be turned off in an organization that requires it.
  • Receipt files and downloads apply the same account checks as the rest of the app (deactivated accounts, role changes, outstanding security steps).
  • Slack, Teams and QuickBooks settings use a person's current role, so a demoted administrator loses access straight away.
  • A phone's notification registration belongs only to the person who signed in on it most recently, so a shared phone no longer receives the previous person's notifications.

Who sees what

  • Department heads and documents: they can open, download, view the audit trail of, rename or delete only documents they sent (plus their own), and can no longer add themselves as a signer. Attaching a document to an onboarding task needs permission to open it.
  • Signers on a document with several signers no longer receive another signer's signature image.
  • A person demoted to employee is no longer an approver. The approver flag stayed on after a demotion (usually an organization's founder who handed over the executive role), which kept their view of the whole organization's pending leave, reasons and private leave types included, and let them approve anyone's leave. Changing someone's role to Employee now clears it (audited), and employees left with it are cleared. Department heads, administrators and executives approve as before.
  • Colleagues' clock-in locations are shown only to department heads (for their people), administrators and executives. Everyone still sees who is clocked in.
  • Company Directory "Admins only" now applies in the app as well as the public API: employees see only themselves in the people list and profiles, and a department head sees full details only for people in their departments. Schedules and timesheets still show names and photos.
  • Calendar subscriptions for the whole organization, direct reports or approvals stop working when the owner's role no longer allows them.
  • On a colleague's onboarding checklist, an employee sees only the tasks assigned to them.
  • Peer and direct-report feedback stays anonymous whatever capitalisation was used.
  • Editing a 1:1 series needs the same rights as setting one up; participants still add meetings and notes.
  • Department heads can no longer edit or delete open shifts filed to other departments in bulk.
  • Custom-domain branding can be changed only by executives, and its privacy, terms and support links must be https addresses. Viewing a custom domain needs administrator access.
  • Organization settings sent to the browser no longer include internal feature switches.

Lists that could widen to everyone

  • Copying a schedule week with no employees selected no longer copies everyone, and its email goes only to the people who got new shifts.
  • A bulk-send row with no email address is rejected; it could reach an unintended employee.
  • An unknown "required for" value on a training course is refused instead of assigning the course to everyone.

Exports and records

  • Downloads of audit logs, sign-in logs, the platform user list, QuickBooks sync history, leave reports, timesheet PDFs, expense report spreadsheets and geofence violations are recorded in the audit log, and text in these files can no longer run as a spreadsheet formula.
  • Billing checkout and portal always return people to BookYourPTO or the organization's own domain.
  • Password setup emails are no longer sent to deactivated people.
  • Import errors no longer show internal details or another platform's raw responses.
  • Mileage distance lookups are rate limited per person.
  • Support tickets can be opened, and files added to them, only by administrators and executives of the organization. The support page shows "not found" to everyone else.
  • Password reset emails always link to your organization's own address, never to whatever address a request claimed to come from.
  • Sign-in through single sign-on never follows a redirect from the identity provider to another address.
  • The app's start-up data no longer includes internal organization switches (such as which people are blocked from the assistant); every member received them before.
  • The public demo no longer sends support tickets or schedule emails.
  • Protection against runaway assistant usage. Each person can have one answer running at a time and send at most 12 questions a minute and 60 an hour, so a burst of parallel requests can no longer slip past the allowance before it is counted.
  • Fewer false "session hijack" alerts. Two tabs (or a network retry) refreshing a sign-in at the same moment was recorded as a HIGH session-hijack violation, often several rows at once. A refresh that loses a race within 60 seconds is now treated as normal; real reuse of an older rotated token is still flagged, once per token.
  • Violation details never hold a token. "Unauthorized access" rows stored the full address, including receipt preview tokens. The address is now recorded without its query string, with secret path segments masked, like the access log. Existing rows are left as they were.
  • Erasing a person's data now includes the assistant. A privacy data-deletion request also deletes their assistant chats, memories and stored files.
  • The assistant flags misuse. Requests for system secrets, other organizations' data, instruction overrides and injection payloads are recorded as "AI Assistant Misuse" violations.
  • Asset categories and certification types could be deactivated across organizations. An administrator who knew the identifier of another organization's asset category or certification type could deactivate it. Both actions are now limited to the administrator's own organization, and an identifier from another organization is answered exactly like one that does not exist.
  • The connector no longer fetches receipt images from a web address. Receipt analysis and upload through the assistant take the file itself. Fetching an address from the server would have let a caller reach addresses inside the hosting network.

Privacy

  • Conversations from before this release are never used to improve Intelligence. Only conversations held after the update, while Help improve Intelligence is on, can be reviewed.
  • Assistant usage signals are kept for no longer than 24 months. A weekly job clears older ones; the conversations themselves are unaffected.
  • The assistant no longer receives the email address of a connected Google calendar account, or the calendar provider's own error text, when it checks calendar sync status.
  • Privacy Policy, Terms of Service and Subprocessors updated to describe BookYourPTO Intelligence in full and the Help improve Intelligence program, with retention limits (usage signals and de-identified or shared conversations: 24 months; analysis copies: deleted within 30 days).

Public API

  • Time entries: GET /api/v1/time-entries?approval_status=PENDING leaves out entries inside a finalized or signed timesheet, matching the approvals queue; approving or rejecting one returns CONFLICT.
  • Leave approvals follow the app's approver rule. POST /api/v1/leaves/{id}/approve and /reject run the same shared service as the app, so a token owned by an employee who only carries the approver flag can decide only leaves assigned to them (403 otherwise).
  • Expense line items include the extra tax. Expense report detail and expense_report.* webhook payloads now carry other_tax and other_tax_label on each line item (additive); total_tax includes it.

Improvements

  • Calendar days are cleaner. Days with leave, holidays, locks or celebrations show just their round coloured icon, as the dashboard does, instead of a bordered square around it.
  • Open shifts can be claimed only in a published week and before they start.
  • A department head filtering a report or violations list by a department they don't manage is told so, rather than shown all their departments under that label.
  • Purchases can carry a named extra tax. Alongside GST, PST, HST, QST and gratuity, a purchase can include one more tax such as a liquor tax, city tax or tourism levy: choose Liquor tax, City tax or Tourism levy, or type your own name. It is included in the purchase total, the report's tax total, Excel exports and QuickBooks. Receipt scanning now reads a separately printed liquor, city or tourism tax into it instead of folding it into sales tax.
  • Expense report numbers work with BookYourPTO Intelligence. Asking about "EXP-2026-0001" now finds that report directly instead of failing and searching for it.
  • Business includes 500 documents a month. The Business plan's document allowance (files and signing requests together) goes from 100 to 500 a month. Existing Business organizations move to the new allowance automatically; an organization with a custom allowance keeps it.
  • BookYourPTO Intelligence suggests what your role can do. The assistant page and panel no longer offer approvals to people who cannot approve, or company-wide reports, sent-document reminders and onboarding status to employees. Suggestions follow the same rules as the sidebar.
  • Assistant sits under Home in the sidebar. Home is now a group like Documents: the chevron folds it, and Assistant is indented below it with an icon the same size as its neighbours.
  • A reached usage limit shows before you type. The assistant page now checks your usage when it opens and when you come back to the tab, so the limit notice appears above the message box straight away instead of only after you ask something. The notice also lines up with the box.
  • A question asked at your usage limit is paused, not lost. If you ask BookYourPTO Intelligence a question after reaching your usage limit, your question is now saved and shown as paused, with when your usage resets. Once it does, press Continue to get your answer. A change waiting for your approval is kept too, and you can still cancel it at any time.
  • Announcements come in three designs, with their own colours. Besides the centred card with a picture, an announcement can be a split panel (a coloured panel beside the text, with the short lines shown as a checked feature list and an optional second button such as "Learn more"), or a small card in the corner that leaves the page usable and needs no dimmed background. Each announcement can set its accent colour and, for the centred card and split panel, its gradient; without them it uses the organization's theme colour as before. The picture's small card now names the feature rather than repeating the full title.
  • Assistant chats get short, clear names. A new chat used to be named after the first 64 characters of the first message, so a question that included a link showed the link in the chat list. A new chat is now named straight away from the message with links, record ids and email addresses taken out ("delete my claim"), and once the first answer is written it gets a short title that says what the chat is about ("Delete draft expense report"), in the language the question was asked in. The new name appears in the header and the chat list without a reload. A chat you renamed keeps your name. Naming a chat counts toward your assistant allowance like any other answer, and is skipped when the allowance is used up.
  • Search your past chats by what was said. The Search box on the assistant page, and a new search box under Recent in the assistant panel, now look through what you and the assistant wrote in your chats, not only their titles, and match your words wherever they appear rather than the exact phrase ("compared leave by department" finds "Compare leave taken by department"). Results show a short excerpt with the words you searched for highlighted, newest first, up to 20. Only your own chats are searched.
  • Ask the assistant to find an earlier chat. "Find my chat about my leave balance" or "where did we talk about my expense claim?" now finds the conversation and answers with a link that opens it. This works in every mode, including Plan, and on every plan, and only ever looks at your own chats.
  • Faster, cheaper, more reliable receipt scanning. Receipts are read by a main AI provider, and a second provider takes over when the first fails, including when its balance runs out. Both use newer models that read every field correctly in our tests at a fraction of the previous cost.
  • The assistant has its own place in the sidebar. "Assistant" sits under Home and opens the full assistant page with your earlier chats. It appears only while you can use it: it is hidden when the assistant is switched off for you, your department or your organization, or paused.
  • Schedule a whole team in one request. Ask the assistant to schedule several people across several days and it does it in one step. It first checks the week, says whether it is a draft or already published and what is scheduled, and always asks before changing a published week.
  • The assistant follows your organization's modules. When Time Tracking & Projects or Expenses is switched off, the assistant no longer offers anything from that module, the same way the sidebar hides it.
  • Attach files to a question. Add images, PDFs or CSV files to a message (up to five, 5 MB each and 10 MB in all) and the assistant reads them, for example a receipt it turns into an expense line.
  • No more double confirmation in the assistant. When a request is clear, the assistant acts according to the mode instead of asking "shall I go ahead?" first: in Ask mode it shows the approval card straight away, in Accept edits and Auto it makes the change (high-risk actions still show a card). It still asks a short question when something is missing, such as which dates.
  • Settings > Intelligence. Executives and administrators choose who can use BookYourPTO Intelligence: on for everyone by default, switched off by department or by person, with people switched back on inside a switched-off department. A Company usage tab shows each person's tokens today, this week and in total.
  • Announcements. Product news can appear once as a window after sign-in, always after the welcome steps and tours. Only the newest is shown, and closing it means it does not come back.
  • Charts can be downloaded as PNG or SVG. Charts keep up to eight series and use round axis steps. Tables in the assistant keep readable columns and scroll sideways only when they need to.
  • Clearer assistant outages. If the assistant is paused or its AI provider is unavailable, people see that it is temporarily unavailable, and that everything else works as usual.
  • Assistant plans. Pro can now take actions with the assistant, not only Business and Enterprise. Each person gets a weekly allowance (Free 600,000 tokens, Pro 1,900,000, Business 3,100,000, Enterprise 9,400,000) and a share of it per session (a seventh), so it lasts the whole week. A session starts with your first question and lasts five hours. Enterprise allowances can be set per organization. Usage shows Current session and This week.
  • Settings → API and the API reference load in their own shape. Instead of a few grey bars, the page shows its tabs, token rows and side cards (and the reference its navigation and endpoint blocks) while it loads.
  • The assistant can do more. New actions include asking to cancel approved leave and reviewing those requests, submitting, approving and finalizing timesheets, uploading receipts, editing per diems and mileage, viewing a document template's versions and archiving or restoring templates, reminding signers, reviewing expiring documents, editing and deleting performance cycles and recording assessments, reading and setting custom profile fields, correcting compensation, job history and other employment records, per-person holiday overrides, and test sends for email, Slack and Teams settings. Two actions that could never work were removed.
  • Connect an AI assistant in one command. npx @bookyourpto/wizard mcp add finds the supported AI coding tools and editors on your computer and adds the BookYourPTO connector to the ones you pick, and shows the in-app steps for desktop AI apps. It stores no password or key: your browser opens to sign in the first time the assistant connects. mcp status and mcp remove show and undo it.
  • BookYourPTO Intelligence, the in-app assistant (beta). Ask HR questions and take actions from a panel on every page, or full page at /assistant. It works only inside BookYourPTO, with your own permissions. Choose how it acts: ask before every change, accept edits, auto (high-risk changes still ask) or plan (read-only). Type / for commands (/new, /compact, /memory, /remember, /mode, /export, /usage and more). It remembers facts you ask it to, which you can see and delete, and keeps long chats short automatically. Attach, paste or drag in a receipt, a note or a screenshot and it reads it — a receipt is attached to the expense it creates; everything you attach is kept on a Files page. It draws charts and makes CSV and PDF files from your data. Reloading the page mid-answer picks the answer back up (or offers Retry if it was cut off), and a change waiting for your confirmation comes back. Skills (/skills) walk through common jobs: a receipt into an expense, planning time off, approvals, timesheets, a self-review, an HR letter. Settings it cannot change, such as passwords or two-factor, get a link to the right page.