# Proactive Employer Messaging Your AI employee does not wait for you to check in. When it finishes a task, hits a blocker, or needs a decision from you, it sends the update itself, straight into the same chat thread you already use with that employee. No separate inbox, no polling a dashboard to see if something happened. By default, a chat thread only moves when you send the first message. Proactive messaging changes that: whatever woke the employee, a schedule, a delegated task from a teammate, or an API call, once it has something worth telling you, it sends a message on its own through a dedicated tool built into every employee, not a separate opt-in feature. The message always lands in the same chat thread you already use to talk to that employee. There is no second inbox to check and no separate notification surface to learn. You keep working in one place, and updates show up there when they are ready, not on a fixed timer. This is different from an approval or choice request, which pauses the employee mid-task until you answer. A proactive update never pauses anything: the employee sends it and keeps going, or has already finished. If a task genuinely cannot continue without your decision, the employee uses a separate request that stops and waits for you, so a blocking question and a passing update never get confused with each other in the thread. The tool carries a priority tier, low, normal, high, or critical, that the employee sets based on how much the update warrants faster reach. A finished report can wait for you to open the app; a blocker that stops the work entirely is flagged to reach you faster. A built-in guard also blocks messages that only narrate intent, so you get results and decisions to make, not a stream of 'I'm about to check on this' noise. ## What triggers a proactive message Any run of an employee can end in an outbound message, not just ones you started. A recurring schedule that kicks off research overnight, a task another employee delegates over to this one, or a request that arrives through the API can all finish with the employee choosing to tell you something. That means you do not have to remember to check on scheduled or delegated work. If there is nothing worth reporting, the employee simply does not send anything. If there is, it reaches you without you asking. ## Why it always lands in the same thread Keeping proactive updates in the same chat thread you already use with that employee means there is nothing new to learn and nothing extra to monitor. You see the update in context with everything else you have discussed with that employee, and you can reply directly to continue the conversation or approve a next step. This also means the feature does not depend on you watching a separate notifications panel. It rides on the chat surface you are already in day to day. ## How the narration guard actually filters messages The guard only looks at how a message opens. If the first line starts with an intent phrase, 'I'll now', 'reaching out to', 'about to check', 'going to', the tool checks the rest of the message for a deliverable marker: a completed-work verb, a link, a blocker, or a question. Find one and the message goes through even though it opened like narration. Find none and the send is rejected, with the employee told to finish the work first and report after. Most real updates carry at least one marker, so the guard stays quiet unless the employee is genuinely only announcing intent. This exists because pure status narration was observed in production: an employee messaging its employer 'Reaching out to check on her progress' with nothing to report. The guard is a cheap regex check, not a model call, so it runs on every send with no added latency. ## Delivery timing by priority The chat message itself is instant regardless of priority. What priority actually changes is how the bell notification escalates to email if you have not seen it yet. A low-priority update waits 2 hours before it can email you, and at most one low-priority email goes out per 6 hours. Normal, the default, waits 30 minutes and shares that same 6-hour email gap. High priority cuts the wait to 5 minutes with a 1-hour gap. Critical priority skips the delay entirely and bypasses every cap, so a genuine blocker reaches your inbox immediately rather than sitting behind the gap. The email step only fires if you have not already read the bell notification, opening the chat marks it read within seconds over the WebSocket connection, so an employer actively working in the app rarely sees the email at all. Two global backstops sit underneath every priority except critical: at most 2 emails per hour and 4 per day across all notification types, so even a chatty employee cannot flood your inbox. If the exact same update text repeats from the same employee, the email side is suppressed for 24 hours even though the bell still rings every time, which catches an employee stuck in a loop re-sending the same 'finished X' status. ## How It Works **One tool, one thread, four priority levels** Every employee has access to an internal tool for sending you a message outside of your own turn. It takes the message text and a priority: low, normal, high, or critical. The employee decides when to use it, typically after finishing a piece of work, hitting something it cannot resolve alone, or needing an answer before it can continue. Before the message goes out, a guard checks the opening line. If it looks like the employee is only narrating what it is about to do, 'reaching out to check on this now', and the message contains no result, blocker, or question, the send is rejected and the employee is told to do the work first and report after. This keeps the thread useful instead of turning into a live process log. Once a message passes that check, it is delivered into your existing chat thread with that employee, regardless of whether a schedule, a delegation from another employee, or an external API call triggered the work. Priority controls how the notification reaches you, so time-sensitive blockers do not sit unread behind routine status updates. ## Use Cases ### Overnight scheduled work You set an employee on a recurring schedule to compile a report every morning. Instead of you opening the app to check whether it ran, the employee messages you directly once the report is ready. ### Blocked on a decision An employee working through a task hits a point where it needs your call to continue, a choice between two approaches, or missing information only you have. It messages you with the specific question instead of stalling silently until you happen to check in. ### Handoff after a delegated task One employee delegates part of a task to another. Once that teammate finishes its piece, it can message you directly with the result, even though you never spoke to it directly to start the work. ## FAQ ### Do I need to turn this on? No. It is built into how employees message you and works out of the box for every employee. There is no separate setting to enable. ### Can an employee spam me with status updates? A guard blocks messages that only announce intent, like saying it is about to check on something, unless the message also carries a real result, finding, blocker, or question. Beyond that, the bell always rings but email delivery is capped: at most one email per priority tier's gap window, 2 per hour and 4 per day across all notifications, and identical repeated updates from the same employee stop emailing for 24 hours even though the bell keeps ringing. ### Will I get a notification outside the app for a critical update? Yes, and faster than any other tier. Critical priority sends the email immediately and bypasses the normal delay and rate caps that apply to low, normal, and high priority messages. The chat message itself always lands in your existing thread with that employee first. ### Does this work for scheduled and delegated tasks, or only tasks I start myself? It works regardless of what triggered the employee, a schedule, a delegation from a teammate, or an API call all can end in a proactive message to you. ### How is this different from an employee asking me to approve something or pick an option? A proactive update never blocks the employee, it sends the message and continues or has already finished. Approval and choice requests are a separate mechanism that pauses the employee's task until you respond, used when the work genuinely cannot continue without your decision. ## Where Proactive Employer Messaging fits Proactive Employer Messaging is part of How they use your computer. Your AI agent operates your browser and desktop apps the way a human would. Navigate web apps, fill forms, click buttons, move files, and complete multi-step workflows across any application. Install the lightweight companion app and your employee gains access to your entire machine. - [How they use your computer](/en/features/automation): Control any browser or desktop app. ## Read the guide - [Guide: Proactive Employer Messaging](/en/guide/work/chat) ## More in Desktop Control - [Browser Automation](/en/features/automation/browser_automation): Most automation tools can read a public web page and stop there. The moment a task needs a real login, a multi-step form, or clicking through an actual workflow, they fall over. Browser Automation gives an employee a real browser on your own machine, driven through the desktop companion app, so it can navigate, click, type, and pull data out of sites the same way you would, including the ones that live behind a sign-in. - [Computer Controller](/en/features/automation/computer_controller): Most AI tools stop at the browser. The moment work lives in Photoshop, Excel, Logic, Final Cut, or a native app with no API, they're stuck. Computer Controller is what lets an employee step outside the browser and operate your actual desktop, real mouse, real keyboard, real screen, through the companion app running on your machine. - [Automatic File Organization](/en/features/automation/file_organizer): Your AI employee sorts, renames, and files documents on your own computer through the Sistava Desktop Companion app: reading what is in a folder, working out what each file is, and moving it where it belongs using the naming convention and folder structure you describe in plain language. File access stays inside your home folder, and every move, rename, or write is recorded in the activity feed so you can see exactly what changed. - [Hands Free Data Entry](/en/features/automation/data_entry_automation): Your AI employee reads records from a spreadsheet, export, email, or PDF and types them into any web or desktop application, including tools with no API. Every field is checked against the source before submission, and anything incomplete or ambiguous is set aside in an exception report instead of entered wrong. Recurring entry jobs run on a schedule with the approval checkpoints you choose. ## Explore - [Every feature](/en/features) - [Hire an AI employee](/en/market) - [Pricing](/en/pricing)