AI Platform Changelog and Product Updates
Track AI platform product updates, including new features, improvements, fixes, and integrations as they ship.
Reliable Workflows for Your AI Employees (2026-09-13)
Choose a complete outcome, keep its progress and approvals in one place, and return to the same saved work when it needs you. Sistava now makes supported employee workflows visible and dependable from start to finish. Choose an outcome from an employee's Work menu or ask for it in chat. The workflow saves the brief, current stage, completed results, and any decision it needs from you, so work can pause and continue without losing context. Reusable playbooks supply the methods behind the work, while each workflow keeps the job's own progress and evidence. When a connected action needs your approval, you see the exact action before it happens. You can keep Ask first in place, request changes, or enable automatic action later for tools you choose to trust.
Live Video Call with your AI Employees (2026-09-03)
Start an audio or video call with an employee directly inside your workspace. You can now start a live audio or video call with an AI employee from chat. Speak naturally, hear replies in real time, and keep the conversation in the same employee thread as your typed messages.
Keep working on your watch (2026-08-31)
Read a direct employee chat and send a short reply from a focused Watch-sized view. Web Chat now has a focused view for an Apple Watch or another very small screen. Choose an employee, read the latest conversation, and send a short reply without opening the full workspace. You can also turn on wrist alerts for a workspace. They tell you that an employee has an update and take you back to the chat, while the employee's message stays private until you open it.
Download Your Full Account Data (2026-08-28)
Request a complete copy from your profile, then download it when it is ready. Your profile now has an Export all my data button. Ask for a complete copy, leave the page while we prepare it, and download it from the workspace as soon as it is ready. Your export stays protected by your signed in account and is available for a limited time. You also receive an email when it is ready, so you do not have to keep checking back.
Bring Knowledge From Other Agents (2026-08-24)
Bring the files and context you built elsewhere into your employees' knowledge. Your Knowledge page can now bring in the work you have already built with other agents, or from any folder you choose. Pick a source through the desktop app, review what will be included, and confirm before anything is copied. Imported material starts private to you and remains yours to review, share, resync, or remove. You can later choose the specific files you want an employee to use as searchable knowledge.
Rate Us Whenever you Can (2026-08-24)
Share a private rating from the header without interrupting your work. You can now rate Sistava directly from the workspace header. Choose a star and your private rating is saved right away, so sharing your experience takes only a moment. If you want to add detail, the feedback drawer still gives you space for a private note. Once you have rated, we will not ask you again.
Work in Your Signed In Browser (2026-08-20)
Your employee can work in a fresh tab inside the browser profile you already use. When you approve browser access through the desktop app, your employee now works in one dedicated tab in the browser profile where you are already signed in. Your existing tabs stay untouched, while the employee can use the accounts and session you have already opened. After the first approval, follow up work keeps using that same browser session instead of opening another browser or asking you to sign in again.
Retry Failed Training in One Click (2026-08-19)
A failed training source now retries with the same setup, nothing to redo. If training on a connected app fails partway through, you can now hit Retry and it starts over with the exact same source and settings you already picked. No reopening the file picker, no reselecting pages or folders. The failed run stays in your history so you can see what happened, while the retry runs as a fresh attempt right alongside it.
Transcribe Uploaded Audio and Video (2026-08-18)
Attach a recording in chat and get a saved transcript back in Drive. Audio Transcriber is now available to every AI employee. Attach an audio or video recording in chat, ask for a transcript, and the employee saves the written result to Drive before continuing with summaries, action items, or follow-up questions. The tool handles common audio and video formats, checks your credit balance before processing, and can add timestamped speaker labels for conversations. It is for recordings you already have. Meeting Attender remains the separate tool for live calls.
Put a Waiting Message at the Front (2026-08-17)
Reorder the messages waiting to send, so the one that matters goes next. While an employee is working you can keep typing, and each message waits its turn in a list above the writing box. Until now they went out strictly in the order you wrote them, so a fifth thought that turned out to be the important one sat behind four others, and the only way to move it up was to cancel the rest and type them again. Now every waiting message has an arrow that jumps it straight to the front, and on a computer you can grab a message by its handle and drop it anywhere in the list. The order you see is the order it sends, and you can change it right up to the moment the employee picks the next one up. The X is still there when a waiting message should not go at all: it pulls the text and any files back into the box so you can rewrite it.
Ask for a Mockup of a Screen (2026-08-17)
Describe a screen or page and your employee designs it as a picture you can share. Create in Content now has a Mockup type, in the Visuals group next to Presentation, Flyer, Infographic, Brochure, Logo and Web page. It asks for two things: the name of the screen, and a short brief covering what the screen is for, the sections that belong on it, and the visual direction you want. Pick the employee to build it and the design comes back as a PNG. It is deliberately a picture and not working code, which is exactly what you want while you are still deciding what something should look like. Nobody has to build anything before you can react to it. Put it in front of a designer, a developer, or a customer, get the reaction, and change the brief instead of changing the build. Like every creator in Content, the card is a short brief rather than an editor, and whatever you typed survives closing it. The finished image comes back through chat and lands in your Drive, where you can mark it up: click the part you want changed for a numbered pin, write your note beside it, and ask for a revised version. A mockup is cheap enough to throw away and redo, which is what makes it useful this early in a decision. One limit worth stating plainly: it is a still image of a single screen. It is not a clickable prototype, and it does not produce a working page.
Ask for Your Data, or Ask Us to Delete It (2026-08-15)
Both buttons on your profile now work. Ask, get an email back, and it is done within 30 days. The two buttons under Your data on your profile page used to tell you to write us an email. Now they open a short form, and a person picks it up at the other end. Asking for a copy starts with choosing what you want: everything your account holds, or just your conversations, your employees and how they are set up, your files and Drive, or your work journals, tasks and activity. If you only need one of those right now, each of those pages can already export itself today without a request. Asking us to close your account works the same way with one difference: we write to you first and confirm the request really came from you, because erasure cannot be undone afterwards. Telling us why you are leaving is optional and stays optional, since exercising a legal right should never be gated behind answering a survey. Deleting the account ends your subscription as part of it, so if you only want to stop paying, cancel the plan and keep the account instead. Either way you get a confirmation in your inbox straight away, and the request is finished within 30 days, usually much sooner, with another email the moment it is done. Copies sitting inside our routine backups roll off within 30 days after that and are never touched for anything in the meantime. Your data is never sold and never exposed to anyone outside Sistava: inside Sistava it is only used to run your workspace and to help our own systems get better at their job.
Memory and Knowledge Now Follow the Role (2026-08-14)
What an employee remembers, reads and writes is now bounded by the access of the person it is working for. Until now an employee's memory and the knowledge it drew on sat outside the permission system. From today they sit inside it. Every knowledge lookup, every fact pulled into an answer, and every note an employee writes back to its own notebook or work journal is checked first against the access of the person it is working for at that moment. That last part matters more than it sounds. An employee is not one fixed identity with one fixed reach: it works for whoever is talking to it. The check is therefore made per action, as the action happens, rather than fixed once when the employee was set up. Ask the same employee the same question as two different people and each of you gets an answer bounded by what you can already see. Trained sources, notebooks and work journals each carry their own setting: shared with the organisation, or restricted to their owner and the people explicitly granted access. Restricted means restricted here. Unlike an ordinary workspace file, an Owner or Admin does not quietly inherit private material by asking an employee for it. They can still manage who has access, but they cannot read through an employee as a side door. Nothing moves and nothing is deleted by this. Narrow someone's access and what an employee recalls for them narrows with it. Widen it again and it widens back.
Your Employees Can Take You Anywhere (2026-08-13)
Ask where something is and your employee opens it for you instead of explaining the way. Ask how to connect an app, where a file ended up, or what a task actually says, and your employee answers the way a colleague would: by opening it. A card appears under the reply naming where you are going, fills up over about five seconds, then takes you there. You stay in charge of it. Click the card to go immediately, or click the cross to stay exactly where you are and keep reading, and it will not try to send you again on the next answer. Only one destination is offered per reply, for the place that matters most, so an answer never turns into a queue of jumps. It covers every page in the workspace, from Chat, Drive, Tasks and Calendar through to CRM, Tools, Training, your dashboards and your settings, and it covers the single record you were just discussing: a task, a project, a sprint, a file, a contact, a teammate, a team. A specific item opens where that item actually lives, with the record in front of you rather than the list it sits inside. Two limits are worth knowing. It can only take you somewhere you were already allowed to go, so it is never a back door to a record you could not find yourself. And it opens a page or a record, not one control buried inside one: ask about a single switch on a busy settings screen and you land on that screen. If nothing in your workspace matches what you asked, you are told so plainly instead of being dropped on the nearest guess.
Mark Up an Image, Send Your Notes Together (2026-08-12)
Pin comments to the exact spots on a picture, then send the whole review with one button. Open an image in Drive and click the part you want changed. A numbered pin appears and you write your note beside that number. Drag across the picture instead of clicking and you get a numbered box covering a whole area. Go round the image marking everything, one remark per mark, the way a design review works: the logo is too dark never has to explain which logo. Then press Send once. That single button is the point of it. Your employee receives the whole review as one round of feedback, reads every remark before it changes anything, and comes back with one revised version that answers all of them. Sent one at a time, you got a version that fixed your first note before it had seen the rest. The marks stay on the picture. Anyone who opens that file later sees which parts already carry feedback and what was said about each, so a second reviewer does not repeat a point you already made, and a resolved remark keeps a quieter mark instead of disappearing. Positions are held relative to the image itself, so a mark sits in the same place in the side drawer, in fullscreen, and on a phone. The marks live in the file's own comment thread, so everything a comment already does works here too: mention a person or an employee on one, or turn a remark into a tracked task in a click.
Highlight Text and Comment On It (2026-08-12)
Select any passage in a document and comment on that exact text. The highlight stays on it. Open a file in Drive and drag across the sentence you want changed. A Comment button appears at the selection, and your note lands in the thread with the passage quoted above it, so you never have to describe which paragraph you meant. The highlight stays on the text. Reopen the file next week and a mark shows every part that already carries feedback. Click a mark to jump to its comment, or click a comment to jump back to its text. A resolved thread keeps a muted mark, so the page stays readable without losing the history of what was raised. If your employee rewrites the file, the marks re-measure themselves against the new text rather than floating over the wrong line. Your employee reads the quoted passage alongside your note, which is what makes short feedback work. Make this shorter finally has something to point at, and this contradicts the paragraph above stops being a guessing game. It works anywhere the file renders as real text: written documents, notes, code, spreadsheets and CSV tables, and Word files. PDFs are the exception, because their text sits inside a separate viewer we cannot attach a mark to, so a PDF falls back to an ordinary comment on the whole file. Images get numbered pins instead.
Start an Article, Deck, or Flyer (2026-08-12)
Content now creates articles, presentations, and flyers, and each one starts in a card shaped like the thing you are making. Pick Article, Presentation, or Flyer in Content and you get that format on screen: an article page with its cover and headline, a title slide with the deck behind it, or a poster with its main visual and headline. Type the headline where the headline goes and describe the rest underneath. For an article that means the topic, the audience, the angle, and the points that matter. It stays a short brief, not an editor. Two fields and a format choice, because the job of this card is to get the request right, not to make you lay anything out. An article can come back as Markdown, a Word document, or a PDF. Whatever you typed survives closing the card, so a stray click never costs you the brief. Hit Create with AI and the employee you picked takes it from there, with the headline, your description, and the format already in hand, then builds the real file: a document, a deck, or a printable PDF. It comes back through chat and lands in your Drive, where you can read it, comment on the exact passage you want changed, and ask for a revision. One thing to be clear about: these longer formats are creators, not managed cards. Unlike a social post, an article or a flyer does not get its own review and approval lane in Content yet. It arrives as a finished file for you to check.
Draft Social Posts in Their Final Format (2026-08-12)
Start a social post in Content and you write straight into the real platform card, then your employee finishes it. Choose Create and then Social post in Content, and you land in the post itself: the profile picture, the handle, the reactions underneath, exactly as it will look once it goes out. Switch the platform at the top and the whole card changes with it, including the character budget, so the limit is visible while you type rather than discovered after you have written past it. Write the finished post yourself or just describe what you want in a couple of lines, then hit Create with AI. Your employee picks it up in chat with the platform and the length already in hand, writes it properly, and saves it back to Content as a real record you can review. Whatever you typed survives closing the card. Because the preview is faithful rather than decorative, you see the line breaks, the truncation, and the picture the way a reader will, which is where most social drafts quietly go wrong. One idea can also become several versions, one per network, each with its own wording and its own picture, all hanging off the same creative so the conversation about it stays in one place. Nothing goes public because an employee decided it should. Proposing a time moves a post into review, and only your approval moves it into the scheduled lane.
Book a Meeting from your Calendar (2026-08-12)
Put a call on your calendar, pick who attends, and your employee joins it without being asked. Getting an employee into a call used to mean catching it at the right moment and handing over the link while the call was already running. Now open your calendar, hit Create, and choose Meeting. Paste the Google Meet, Zoom, or Microsoft Teams link, say when it starts and how long it runs, and pick who attends. The meeting sits on the calendar next to everything else you have planned, and when the time comes your employee walks in on its own. It listens, speaks when someone says its name, and writes up the summary afterwards, whether or not you are in the call yourself. That is the part worth booking in advance: a weekly client call or a standup no longer needs you present just so it gets recorded and understood. Meetings you were already invited to through your own calendar app show up on the same grid, so the week reads as one picture instead of two. Drag a meeting you booked here to a new time and your employee's arrival moves with it, because the card on the calendar is the real record rather than a copy of one. Create also places a task, a routine, or an email straight onto the day you clicked, which is what turns the calendar from somewhere you look into somewhere you plan.
Add Deals and Companies Yourself (2026-08-12)
The create button in your CRM now matches the tab you're on: contact, deal, or company. Your CRM has three views, but until now only contacts could be added by hand. A deal in the pipeline or a company on the list meant asking an employee to open it for you. Now the button in the corner follows whichever tab you are on. On Deals it reads New deal and opens a short form: the deal name, the stage it starts in, its value in US dollars, a close date, and the company and contact it belongs to. On Companies it reads New company and asks for two things, the name and the domain. Either way the record appears in the pipeline or the list straight away, no refresh and no waiting on an employee to act. None of this takes the work back from your employees. They read and write the same records, so a deal you opened by hand is one they can move forward, re-price, or close as the conversation progresses, and a deal they opened is one you can correct without leaving the board. The form keeps the hand-off in its footer too: fill it in yourself, or pass it to an employee from the same panel. Deals, companies, and contacts stay separate records with separate access, so sharing one of them never quietly shares the others attached to it.
Delete Files From Your Drive (2026-08-11)
Remove a file you don't want, get the storage back at once, and change your mind for 14 days. Until now a file could only be emptied, never removed, so old drafts, dead screenshots and pictures piled up and kept taking space you were paying for. Open any file and Delete now sits beside Download and Share. Pictures, videos, documents, presentations, spreadsheets and anything you uploaded can go. What stays is your employee's own notebook and work logs. Those are how it remembers what it knows and what it did, so removing them would quietly damage the employee rather than tidy your Drive. They show the button greyed out with the reason on it, so you never have to guess whether something can be removed. Your storage allowance comes back the moment you delete, not when the trash finally empties. That is the part that matters if you were close to your limit: the space is usable again immediately rather than two weeks later. Deleted files go to Trash and stay recoverable there for 14 days. Restore one and it goes back where it was. Leave it and it is purged for good once the window closes, the record and the stored file with it, so a delete you meant is a delete that actually happened. If a file you were looking for has gone, Trash is the first place to check.
Act on a Team From Its Dashboard (2026-08-10)
Every part of a team's dashboard now links to the place where you can do something about it. A team dashboard was good at telling you how a team is doing and useless at helping you do anything about it. You read that three pieces of work were blocked, then went hunting through tabs for the board they were sitting on. Every part of it is now a door. Blocked work opens the task board. The roster opens members, and opens hiring when the answer is that the team is short. The sprint opens the work inside it and the chat where its goal gets changed. Goals and metrics open your business page, which is where they are actually edited rather than merely displayed. Files, routines and activity each open their own tab. The rule behind it runs across the dashboards: the overview says what, and the page it links to says why and lets you change it. Nothing on a dashboard should be a dead end that leaves you navigating by memory.
Open the Detail Behind Any Number (2026-08-10)
Every part of your company overview now links straight to the page that explains it. Your company overview answers how things are going. Until now that was where the answer stopped, and finding the detail behind a number meant already knowing which page owned it. Every section now carries a link to the page that holds the whole story. Your spend opens credit usage. Your activity opens the full log. Your connected tools open channels. The number stays where you can glance at it and the page behind it is one click away rather than a search. A new Controls row sits at the bottom, and it answers a different question: not what your team did, but how it is set up to work. Five tiles, each a door into the setting behind it. The capability level your employees run on, whether Credits Saver is on, whether quality checks score work after each turn, how much of your spending limit is used, and how many channels are genuinely live rather than half connected. A workspace with no limit set reads None set instead of showing an empty bar that looks like plenty left. The pattern is the same across the dashboards: the overview tells you what, the page behind it tells you why and lets you change it.
Turn Briefs Into Real Decks (2026-08-10)
Your employees can now create detailed, editable presentations from a brief, document, or conversation. Give an employee a topic, a report it already wrote, or a spreadsheet of numbers it already pulled, and it now builds the presentation itself rather than handing you slide text to paste somewhere. It plans the structure before it writes anything: how many slides, what each one covers, and what order builds the argument. That is the genuinely hard part of a deck and the part a chat window usually skips. Then it writes the content, keeping bullets short enough to read at a glance instead of becoming a script, and puts the detail that does not belong on screen into the speaker notes. What comes back is a real .pptx, not a wall of markdown you reformat yourself. Open it in PowerPoint, Google Slides, or Keynote and adjust from there. It is produced by the same rendering engine behind the spreadsheets, documents and PDFs your employees already hand you, so the file arrives properly formed rather than as a broken export, and no employee has to write and run code to make it. Because it is a real file it lands in your Drive with everything else. Ask for a revision later and the employee works from the deck it made rather than starting a new narrative, so the presentation stays consistent with the report you reviewed.
Create and Review Company Content (2026-08-10)
Content gives social drafts a dedicated review space and makes the formats coming next visible. Content opens beside CRM in the company workspace, and it exists because drafts otherwise live in a notes app, a half finished document, and a chat message to whoever should review them. That is how good drafts quietly go missing. What it does today is give social drafts a home. Saved LinkedIn and X posts render as real platform cards rather than plain text in a list, so you read a draft the way a follower will see it, line breaks and truncation included, and you can flip one saved draft between its LinkedIn and X previews to check it survives both. It also makes the direction visible rather than hiding it. Articles, presentations and flyers are listed in the workspace now and become available as each editing flow arrives. That is deliberate: a roadmap you can see beats a menu that silently grows, and it lets you tell what this workspace is for before every format is finished. Nothing here publishes on its own. Content is where company work gets created and reviewed before it goes anywhere.
Find Work From Anywhere (2026-08-09)
Global Search now finds workspace pages, people, work, files, and CRM contacts from the persistent sidebar. Press Command K, or Control K on Windows and Linux, from any workspace page and jump straight to the page, employee, team, marketplace candidate, task, project, sprint, Drive file, or CRM contact you need. Every result says what it is, so a task, a file, and a contact with similar names stay easy to tell apart, and every result is a real destination: choosing one opens that record with the right thing already selected rather than handing you a list to search a second time. The page you are on gives the ranking useful context without narrowing what you can find. On Tasks, work leads. On Drive, files lead. On CRM, contacts lead. Results from the scope you are already in get a preference when matches are otherwise equal, and nothing is filtered out, so you can still jump from a task straight to a contact. Worth knowing what it is not. This matches names and metadata, a task title, a file name, a contact's name and email, not the words inside a document or a chat message. Full-text search over your files is a separate capability rather than something this quietly does in the background. Results are also capped per type so one crowded category cannot push everything else off the list. Each resource is checked against your own access before it comes back, so a record you are not allowed to see never reaches the list at all. It is the same index your employees navigate by, which is why anything you can find by typing its name, they can find by hearing it.
Control What Your Employee Shares (2026-08-09)
Employees now protect restricted workplace information and can request permission for one specific answer. Your employees keep using organisation context to do the work, while treating private, confidential and role-restricted information with care. They sort what they know into four kinds: organisation-shared business context anyone in the workspace can draw on, role-restricted detail tied to a responsibility such as finance or people operations, private information from a one-to-one conversation or about a specific person, and confidential information that needs a deliberate decision regardless of who is asking. A workspace role of Owner, Admin or Member describes access to the platform, not automatic access to every private fact an employee has picked up along the way. A job title someone mentions is context for how to work with them, never a key. When someone asks across one of those boundaries, the employee says plainly that it cannot share the detail. It does not pretend to have forgotten, it does not invent an answer, and it does not name who does know. With the requester's agreement it can then ask permission for that one answer. It never forwards the conversation: it reduces the request to the smallest possible disclosure, one figure or one fact, and sends that alone to the person entitled to decide. They see the proposed disclosure and its boundary in a notification, with no transcript and no link into the requester's chat, then approve or reject. Approved, the employee shares exactly that. Rejected, it says it cannot share the information and does not reveal that anyone was asked. One approval is never blanket access: yes to a single number is not yes to the conversation it came from.
Share a single file or contact, not your whole workspace (2026-08-09)
You can now share one Drive file, Drive document, or CRM contact with someone outside your workspace, without giving them access to anything else. Open a Drive file, a Drive document, or a CRM contact and share it directly with any signed-in Sistava account, including someone in a completely different workspace. Sharing by email is enough; there is no invite to your organisation attached to it, and being handed one file never makes anyone a member of your workspace. What the recipient gets is a protected page holding that item and nothing else. From there they can accept it into one of their own workspaces, which creates their own private copy rather than a live window into yours. That distinction is the whole design: an external recipient reads what you shared, and cannot walk from it into your files, your contacts, or anything else you own. You keep the controls. Choose the role for each person you share with, set an expiry if the access should not be permanent, and revoke at any time. Revoking stops further access immediately and never reaches into a copy someone already accepted, because deleting other people's records to tidy up ours would be worse than the problem it solves. Each item is shared on its own terms. A contact you shared does not carry its company or its deals with it, and a file shared from a project does not open the project. If you want someone to have two things, you share two things.
Share Drive Resources With Your Team (2026-08-08)
Give specific workspace members access to files and choose whether they can view or manage each resource. Drive files can now be shared deliberately instead of relying on everyone in the workspace being able to see everything. Open a file, choose a person, pick what they may do with it, and share it from the Resource Access panel. They get an in-app notification that takes them to the file. Access levels are separate on purpose. Read-only opens the file. Can view and comment lets someone leave feedback on it without changing it, which is what most review actually needs. Can edit also allows deleting, because anyone able to rewrite a file to empty gains nothing from being refused the delete, and an emptied file that still looks real is the worse outcome. General access for the file is a separate control from the people you named. Leave it open to your organisation, or restrict it so only the owner, your admins, and the people you granted can open it. You can copy an internal link for anyone who already has access, and remove a person's access later without touching anyone else's. This is internal sharing, scoped to your organisation, and it never makes a file public. Restricting a file is also additive and reversible: a file with no policy keeps the visibility it always had, so turning this on for one file never quietly hides the rest of your Drive from your colleagues.
Comment on CRM contacts (2026-08-08)
You can now leave comments directly on a CRM contact so notes stay attached to that person. Every contact in your CRM now has its own comment thread. Leave a note on a person and it stays with that person, instead of getting buried in a task that closes or a chat log that scrolls away a week later. That matters most for the things that are true about a relationship rather than about one piece of work. They only take calls on Thursdays. Their finance lead is the one who actually signs. They churned once before and came back. None of that belongs to a task, so a task is exactly where it used to disappear. It is a normal comment thread, so it behaves like the ones on tasks and files. Mention a colleague and they get a notification that opens the exact comment. Mention an employee and the comment arrives as a real work request it acts on and answers in the same thread, rather than a notification it merely saw. Turn a comment into a task in one click when a note turns out to need doing, and the new task carries the original text so the request keeps its context on the board while the thread stays put as the record of the discussion.
Request an access review (2026-08-08)
Any non-Owner member of an organisation can now ask every Owner to review their current permissions in one action. If your role or your access no longer feel right, you can flag it yourself instead of waiting for someone to notice. The button sits in organisation settings next to your own effective access and the full permission matrix, so you can see exactly what you hold today, and what each role can do, before you ask anything. Quite often that answers the question on its own. Sending the request notifies every Owner of the organisation at once. There is no queue to join, no approval workflow, and nobody outside the Owners is told. Clicking it twice in a few minutes does not flood anyone's notifications either. The request changes nothing by itself, and that is the point. It is a signal, not a mutation. An Owner opens your membership, checks your permissions against what your work actually needs, and adjusts them if a change is warranted. It is not an escalation, not a complaint, and not a request for more power. Owners cannot send one, because they already hold full organisation access and have nothing to have reviewed. A review also has a ceiling: an Owner can adjust the permissions that are theirs to delegate, but control of the workspace itself, renaming or deleting it, stays with Owners and can never be handed over through a review. Permission drift is almost never noticed by the person who caused it, so letting the person who hits the wall raise their hand turns a periodic audit somebody has to remember into something that happens the moment it matters.
See Sprint Progress At A Glance (2026-08-07)
Follow your sprint's goal, weekly rhythm, and completed tasks without giving up board space. Your task board and your team chats now keep the current sprint visible in one compact strip instead of behind a tab. You see the goal the team is aiming at, how much of the work is finished as a progress ring, and where the week stands on a fixed seven-day tracker, all before you open anything. The combination is the point. A completion percentage on its own tells you very little mid-week: half the work done on day two is fine, half done on day six is not. Putting the goal, the progress, and the day of the week side by side is what makes the glance worth taking. Open the full sprint controls from the same place when you want to change something, whether that is sharpening the goal, moving work in or out, or closing the sprint with its review. Sprints stay optional, and plenty of good weeks are just a board worked in order. When no sprint is running, the same space shows how your loose tasks are going and gives you a clear way to start one at the point the work actually needs a shared goal and an end date, rather than nagging you into one first.
Know When A Schedule Is Saving A Look (2026-08-07)
See whether a schedule is on pace, checking less often because work is quiet, or waiting normally on standby. Schedule cards now make normal quiet-time behavior readable at a glance. An employee that has woken several times and found nothing waiting starts checking less often on purpose. The card now says Pacing itself, instead of leaving that healthy efficiency looking like something had gone wrong. Three states are visible before you open a card: running on pace, pacing itself because the work is quiet, and waiting normally on standby. Real schedule problems stay marked separately, which is the part that matters. A genuine failure should never hide inside sensible behavior, and sensible behavior should never be read as a failure. If a faster check matters right now, choose Use normal cadence. That restores the next check to the cadence you set without editing the schedule itself, so you get one prompt check and nothing to remember to put back afterwards. It is worth knowing why the backing off happens at all. Waking an employee costs credits whether or not there was anything to do, so slowing down during a quiet stretch is the cheapest thing it can do on your behalf. It picks its normal rhythm back up on its own as soon as work appears.
Start A Job Before You Know Every Detail (2026-08-07)
Choose a job from your employee's first card and it now collects the one thing it needs to begin, instead of treating your choice as a dead end. Choosing a job from an employee's first chat card is now treated as a real instruction rather than a menu click that leads nowhere. Before this, picking one could dead-end on something the employee needed and did not have, which is a poor way to spend your first two minutes with a new hire. Now, if it needs one fact to do the work properly, your website or a short description of who you sell to, it asks for that next and says what the answer unlocks. One question with a reason attached, rather than a settings form to fill in before anything can start. Employees also check what a job needs before committing to it, which changes the order the work happens in. Research and drafting come first, and you are asked to connect a mailbox at the point something is actually ready to send. Being asked for a connection you do not need yet is what made a first session feel like setup instead of work. Two things still stop and wait for you on purpose: anything that costs money, such as a paid data pull, and anything that leaves your company. Those ask for your approval rather than being decided by an employee keen to get started.
See Every Employee's Files In One Place (2026-08-06)
Company Drive rolls every employee's Drive into one company-wide view you can browse and search. An employee's own Drive is where its work lands: notebook entries, uploads, transcripts, deliverables. That's useful when you know which employee produced something. Company Drive is for when you don't. It rolls every employee's Drive into one company-wide tab, organised by team, with each team holding its members' Drives inside it and a Bench group for anyone not on a team right now. Employees who have left drop out of the view entirely, so the tree shows who is actually working here today rather than everyone who ever passed through. Nothing moves and nothing is duplicated. It is a read view over the same files: open a report here and you are looking at the exact document its owner has. A toolbar toggle switches between Work files, meaning the documents, images, video and screenshots an employee produced, and All files, which adds its notebook and work logs for when you want the reasoning behind a result rather than just the result. Files a person uploaded, chat attachments and training documents, belong to no employee, so they sit in their own Uploads branch instead of under somebody's name, and can be filtered by project. There is also one Download drive action that zips every active employee's files into a single archive, fetching the contents rather than listing links, so what you get is a real snapshot you can keep offsite or hand to someone outside the product.
Have An Employee Send From Your Own Inbox (2026-08-06)
Connect your own Gmail or Outlook and an employee sends and replies from your real address, with no sending limit. Every employee already has a free built-in email address it can send and receive from the moment it's hired. That's convenient, but it sends as the employee, not as you, and its sending is capped by plan because the address runs on shared infrastructure. Connect your own Gmail or Outlook instead, and an employee sends and replies from your real inbox, under your name, exactly as if you'd typed the reply yourself. It goes out through your actual account, so it lands in your Sent folder, threads correctly, and replies come back to you rather than splitting the conversation into a second thread. There is no monthly sending cap on a connected inbox, on any plan. The connection lives on your workspace rather than on one employee, so you authorise it once with the same one-click flow every other app uses, and every employee you have hired can use it. No app password, no mail server, no host and port to get wrong. Disconnect it and they stop being able to send from it immediately. Reading is instant: an employee can open threads, search, and summarise without asking. Sending is deliberately different. Every send, reply, forward and scheduled send holds for an approval card first, even where you have set that app's other actions to run automatically. A sent email cannot be recalled and it goes out wearing your name, so that stays a fixed rule rather than a preference someone can switch off later. We also never mirror your mailbox into the platform; what we keep is a record of what an employee actually sent. Both mailboxes can be active at once, and the employee does not guess between them. When a request is open-ended enough that the context does not settle it, it asks which address you meant.
Give An Employee Its Own Webhook To Trigger (2026-08-06)
Point an employee at any webhook and it becomes a tool the employee can trigger on its own, mid-task. Point an employee at any outbound webhook, your own automation, a Zapier or Make scenario, an internal service, anything that accepts an HTTP call, and it becomes a tool the employee can use on its own during a task, not something you fire manually from a dashboard. Set it up once with a URL and an optional auth header. Each webhook becomes exactly one send tool scoped to that URL and attached to the team, so add three and an employee sees three separate send tools and picks the right one for the task. You can enable or disable one for a single employee, overriding the team. At runtime the call is a POST with a JSON body, your auth header passed through as you gave it, and the employee waits up to 30 seconds. It gets back the status code and a slice of the response, enough to know the call landed. If it fails, on a bad URL, a timeout, or a blocked address, the employee tells you the webhook failed rather than quietly treating the task as done. By default an outgoing call pauses and shows you the exact payload before it goes, the same as any other action that reaches outside the platform. Turn that off for a specific webhook once you trust it. Two honest limits. The URL is checked when you save it and rejected if it does not answer, and calls to internal, private, or cloud metadata addresses are refused, with redirects not followed, so a public-looking URL cannot hop somewhere internal. And it is one-way by design: this triggers something, it does not query it. Having an employee read data back from an API and reason over it is a separate capability.
Get Back Real Files, Not Pasted Text (2026-08-06)
Ask an employee for a spreadsheet, document, deck, or PDF and get back a real downloadable file, included on every plan, Free too. Ask an employee for a budget and get back a real .xlsx with working formulas and number formatting, not a table pasted into chat that you have to rebuild yourself. Ask for a proposal and get a .docx. A deck comes back as a .pptx with real layouts, charts, images, tables and speaker notes. A data pull comes back as a .csv, and anything meant to go outside the company as a .pdf, from a plain report to a fully designed page with proper print typography. It edits as well as creates. Hand an employee a file you already have and name the change, and it works on that file instead of starting over, so the formatting and content you didn't ask about stay as they were wherever the format allows it. PDFs stay PDFs, regenerated from their readable content. None of this runs through code an employee writes on the spot. A structured rendering engine turns the content plan into the real file, which is why what you get opens properly rather than arriving as a broken export. It is a built-in capability of every employee. Nothing to connect, no separate app to open, no account to authorise, and it is included on every plan, Free included. The file lands in your Drive alongside everything else, ready to download, share, or hand back with a comment asking for one specific change.
Your Employee Checks Real Vendors First (2026-08-06)
Ask for a vendor and your AI employee checks a network of real, vetted businesses first, before the open internet, and you can get your own business listed there too. Ask any AI employee to find a bookkeeper, a contractor, or a service provider, and it checks the B2B Marketplace first, a curated network of real, vetted businesses, before it ever searches the open internet. Nothing to connect or switch on: every employee carries this the same way it carries web search, and it falls back to a general web search only when nothing in the network fits. That ordering is the whole point. A match here already cleared a real qualification bar, so it arrives with a business name, category, region and website rather than a summarised guess from a page somebody wrote about themselves. There are two ways in. A Sistava customer can earn a listing with no monthly fee after being active for a stretch, reaching the Founder plan or higher, and passing a real business review. A business outside Sistava can apply for a sponsored listing, reviewed first and charged only after acceptance, and clearly labelled as sponsored. Neither a plan nor a fee skips the review, and sponsorship never makes a business a better match than it is. The network works both ways. When an employee recommends your business you get a notification naming which employee it was and what they were asked for, and when somebody clicks through to your website from your listing you hear about that too. A directory tells you a name; this also tells the business that somebody was looking for them.
Talk to Your Team About the Sprint (2026-08-05)
Open a sprint and leave a comment on it. Your team leader reads it and answers right there. A sprint is the one thing you and your team are both working on all week, and until now there was nowhere to say anything about it. Asking in chat meant explaining which sprint and which goal first. The sprint now carries its own conversation at the bottom of the sprint panel. Ask whether the goal is still right, push back on it, flag a blocker, or ask whether the week is going to land. You do not have to mention anybody: a plain comment routes to the team leader who owns that sprint, the employee that set its goal and is accountable for it. Mention a different active employee on the team and that person wakes instead, which is what you want when the leader is not the right one to answer a detail. The reply is not a notification sitting somewhere. The comment wakes the leader through the same path as a chat message, so an answer streams back into the thread in seconds rather than waiting for the sprint's next scheduled check-in, and a delivery that fails on a hiccup is retried rather than silently lost. Everything stays attached to the sprint, and every comment deep-links back to the team's board with that sprint selected, so anyone opening the thread later lands where the work is. Use the sprint thread for scope and planning; a comment on a task inside the sprint still goes to whoever is accountable for that one piece of work.
Sprints, Rebuilt Around One Goal (2026-08-05)
Name a stretch of work, give it a goal, and your team runs it and wraps it up with you. A sprint is now a simple thing: a named stretch of work with one goal you agree together. Your team lead can open one when the work is worth naming, delegate it and run it on its own schedule, keep it pointed at the outcome, and close it with a written account of what shipped and what did not. That last part is the reason to bother with sprints at all. A stretch of work that closes with an honest review tells you something a rolling task list never does: whether the goal was the right one, and whether a week of work actually moved it. Sprints stay optional, deliberately. Plenty of good weeks are just a board worked in order, a team with no sprint runs its board exactly as before, and your team will not push you into one you did not ask for. Once a week, if there is genuinely something to decide, your lead raises it with you: what the next stretch should aim at, or how the last one went. Goals are editable again, so a sprint can be sharpened mid-week instead of staying wrong until it ends, and the dashboard now shows what is actually running rather than a stale picture of it.
Fix: Finished On Time Always Showed n/a (2026-08-05)
The Finished on time figure on your company overview sat at n/a even when your team had closed work that week. It now shows the real percentage. The figure only counted work that carried a deadline someone had typed in by hand, and almost nobody types one. So it read n/a for nearly every company, sitting right next to a line saying how much had just been finished, which made the panel look broken rather than empty. It now measures against the deadline your team was actually working to: the end of the sprint the work sat in. That is the real commitment in most weeks, and it is one nobody has to remember to set, because opening a sprint sets it for everything inside it. A date you set yourself still wins. Put a deadline on one specific piece of work and that is what it is judged against, sprint or no sprint. Work that belongs to no sprint and carries no date of its own has nothing to be measured against, so it stays out of the figure entirely rather than quietly counting as early. That keeps the percentage honest in the direction that matters: it can read worse than a flattering number would, and it will never read better than the truth.
A shared calendar for you and your employees (2026-08-05)
See every task with a due date, yours and your employees', on one company-wide calendar. Any task with a due date now shows up on a shared Company Calendar, reached from the workspace icon rail, with day, week, and month views. It doesn't matter whether you set the date or an employee scheduled it while planning its own work: it's all on one timeline. There is nothing to keep in sync, because the calendar doesn't store its own events. It reads due dates off the tasks already on your board, so setting or moving a date in either place shows up in both without anything being entered twice. Click a task on the calendar and it opens the same drawer you already use on the board, with the assignee, the status, and the history. The calendar is a second way to browse the same work, not a separate record to keep straight. Pick a day and you can create a task straight onto that date with the employee already filled in. The point of one shared grid is that it holds both halves of the week. Most calendars show what the people are doing, while whatever the AI side has planned lives in a system nobody opens. Here your work and your employees' work sit on the same timeline, which is the only way to tell whether a day is genuinely full.
Group Tasks And Files By Project (2026-08-05)
Create a project for a client or initiative and your board, your Drive, and your employees all understand what it's for. Create a project with a name and a clear purpose, then attach tasks and files to it as they come in. Select it in the switcher and the task board and your Drive narrow to just that work; clear the filter and everything is back. Pin the ones you're actively running so they sort to the top, and archive the ones that have wrapped: an archived project keeps its tasks and files exactly where they are, it just drops out of the picker and stops counting against your plan's active-project limit. The reason to file something is not tidiness. A task inside a project carries that project's name and purpose with it, so the employee working it knows which effort it serves rather than only what the one instruction says. Past two or three employees, that is the difference between a board that reads like a company and one that reads like a single overloaded to-do list. Your employees can manage the list too: read it, create a project, update a name or purpose, pin one to the top. Archiving stays yours to approve, because it changes what everybody else can see. Nothing is forced into a project. Existing tasks and files read as Unassigned and stay visible, so there is no cleanup pass to do before this is useful. Two limits worth knowing: only tasks and Drive files carry a project today, so conversations, mailbox and CRM stay shared across the whole workspace, and a project is a filter rather than a wall. An employee working inside one project is not fenced off from what it knows about another.
Bookmarks: Your Own Sites, Inside Your Workspace (2026-08-04)
Add any website by its link and it opens right inside your sidebar, so you stop bouncing between browser tabs to get your work done. Click Add Bookmark at the bottom of the sidebar, paste a link, and it shows up as its own row with the site's name and icon filled in for you. The name comes from the domain rather than the page title, so a bank login page that calls itself Sign In still appears as something you recognise, and adding the same site twice numbers the second row so the two stay apart. Rename either one afterwards. Click the row and the site renders right there, inside your workspace, with no new tab to lose track of. Drag to reorder, rename or remove any of them at any time. Bookmarks are yours alone. They are separate from the apps you connect for your employees, your employees cannot use them, and your sidebar never changes what a teammate sees in theirs. It is a personal command centre rather than a shared setting somebody has to agree to. Some sites, typically anything you sign in to, block themselves from being displayed inside another app as a security measure of their own. That refusal never reaches the browser as an error anyone can catch, which is why a link like that would otherwise open a blank frame with no explanation. We check for it when you add the link, tell you upfront, and give you a one-click button that opens the site in its own tab instead. The link is also checked as a safe public address before we ever fetch it, and every redirect it follows is checked the same way.
Every App Now Tells You What It Can't Do (2026-08-04)
Open any connected app and you'll see the full list of what it can do, and just as importantly, the things it can't. No more finding out by failing. Connecting an app used to be an act of faith. You'd link LinkedIn, ask your employee to check the comments on your post, and only then learn that LinkedIn doesn't allow it. The failure looked like your employee being incapable, when the limit was on the other side of the connection entirely. Now every app carries its own honest description. Open one and there's a Capabilities section: the limits first, in plain language, then every action the app genuinely supports and what each one actually does. Limits come first deliberately. What a tool cannot do is the thing you need to know before you plan work around it, and it is exactly the part nobody publishes. Finding out by failing costs you the task, the time, and some trust in your employee. The two halves come from different places, which is worth knowing. The list of abilities comes straight from the provider, so it stays current on its own as they add and remove things. The limits are researched and written by us, because no provider maintains a public list of what their tool refuses to do. This starts with the social apps, where the gap between what people reasonably expect and what the platforms permit is widest, and expands from there.
Recurring Tasks Are Now Routines (2026-08-04)
The same idea with more behind it: a routine has its own instructions, its own cadence, and its own run history, instead of being a card that copied itself. A recurring task could only do one thing: come back, filling your board with rhythms instead of real work. Now there are two clear things. A task happens once and lives on the board. A routine is a rhythm, with its own instructions, its own cadence, and its own run history. Every run sends the employee to do the job again rather than replaying the last result, so a weekly competitor report reflects this week's competitors instead of last week's cached answer. Each run is logged on the routine's own card with its status, and a card only reaches your board when the work genuinely needs you: it wants your review, it spans more than one session, or it is waiting on something only you can give. That split keeps the board meaning one thing, work that wants a human, while the routine keeps the full record of everything it did. Change the cadence, the instructions, or who it belongs to and it takes effect from the next occurrence forward. Nothing already completed is rewritten. Two things worth knowing. Every employee gets a default board-check routine when it is hired, and your working-hours setting is what controls that one; add your first custom routine and the default steps aside so the same check does not run twice. And a routine stays quiet until a real person has messaged that employee or given it a task at least once, so a brand-new hire does not start working on a timer before you have said anything. Run now counts as that first contact, which makes it the quickest way to confirm a new routine actually works. Everything you had repeating moved across automatically, on the same schedule, with the same person on it.
Block Time, Not Just a Deadline (2026-08-03)
A task can now run from one time to another, or across several days, so the calendar holds your own working day and not only your team's to-do list. Until now a task was a single moment: it was due, and that was all it could say. That works for your team, who pick work up and run it, but not for you. You block out an afternoon, you reserve Thursday morning, you give a piece of work three days. So a task can now carry an end as well as a due date. Drag the edge of anything on the calendar to give it a length, drag across days to make it span them, and move it anywhere: it keeps the length you gave it. Dragging moves the real record rather than a copy of it, so what you see on the calendar and what the board holds never drift apart. Leave the end off and nothing changes. A task with only a due date behaves exactly as it always has, which matters because most of what your employees do is a deadline rather than a block of time. The point of all this is that the calendar can finally hold your own working day next to your team's, instead of being a list of their deadlines. Whether Thursday is actually free is a question you can only answer by seeing both.
Training Now Reads Word, Excel And PowerPoint (2026-08-03)
Word, Excel and PowerPoint files you trained on were accepted and marked done without ever being read. They are read now, and any file that fails says so. If you uploaded a .docx, .xlsx or .pptx to train your team, it was accepted, it showed as Completed, and your employee learned nothing from it. Only PDFs and plain text files were actually being read. Nothing on screen gave that away, so the only way to notice was to ask your employee about the document and be told it did not exist. Every format we accept is now read by us, directly, before anything is stored: Word, Excel, PowerPoint, PDF, OpenDocument, RTF, text, CSV, JSON and XML. If a file cannot be read for a real reason, such as a scan with no text in it, that file is marked as failed with the reason next to it and the source shows how many items did not make it. A green tick now means what it says. If you trained on an Office document before today, train on it again. It is worth checking: ask your employee what it knows about the file.
Files Now Sort Themselves Into Groups (2026-08-03)
Your files page was one long wall of everything. It now groups files by where they came from, newest first, with the work your team produced at the top. Finding one document meant scanning every file you own, mixed together: a report your employee wrote sat next to a screenshot it took while working and a note it keeps for itself. There was no way to tell at a glance which was which. Files are now grouped by where they came from: work your team produced for you, pictures and videos it made, screenshots it captured, files you uploaded, and the notes your employee keeps for its own use. Nothing gets filed by hand. Each file is tagged with its origin at the moment it is created, which is why a written report and a Markdown note both land in work files while a screenshot taken during a browser run lands in screenshots regardless of what it happens to show. Within each group the newest is first, and if your employee filed something into its own folder, that folder still shows as a folder rather than being flattened away. The folder view and the download you get from this page use exactly the same groups, so the names never change on you between browsing and exporting. Grouping by origin rather than by file type is the deliberate part. You almost never go looking for a .png. You go looking for the screenshot your employee took while it was stuck, or the report it wrote last Tuesday, and those are two different piles.