Every night — a copy of all data in your folder

The data in the platform is months and years of a company’s work: clients, deals, tasks, correspondence, files. Now every night the platform archives all of it by itself and puts the archive into a folder you choose — including a network folder, separate from the machine it runs on.

The folder keeps the latest copies, so the disk does not fill up.

There are no access keys in the archive — they stay only on the platform’s machine. A copy on a shared drive should not open access to your services.

And a small thing that saves you from an unpleasant surprise: the platform checks that it can write to the folder the moment you set it. You will not find out that there were no copies on the day you need one.

Task updates by email for people who are not on Telegram

The Telegram control panel is handy, but not every employee uses it, and some companies discourage messengers on work machines. Now people who have not connected Telegram get updates on their tasks by email at their work address.

The email is sent from your company’s own mailbox, with no third-party mailing service.

Each person gets an update through one channel — Telegram or email, never both. Several comments in a row arrive as one email, not a pile.

What matters to us is that people learn about their tasks rather than lose them among notifications.

Money in the CRM: a deal closes only with paperwork, and the subscription fee reminds you itself

The costliest sales mistakes happen after the negotiation. A deal is marked closed, but there is no invoice. A client pays monthly, and someone remembers the invoice on the twentieth. We have closed both gaps.

A deal closes only with the paper that matches its payment form. Paying by invoice — you need an invoice. By contract — a contract. Cash on delivery — a waybill. Prepayment — a “Paid” mark. If you try to close a deal without the paper, the platform tells you exactly what is missing.

Contract, invoice and certificate of completion live in the deal card itself: type, number, date, amount and file. The file sits in the platform’s storage, not as a link to someone else’s drive, so the card does not break when a folder is moved. Document numbers are issued by the platform, and an issued number is never reused. A number from the counterparty’s paper is recorded exactly as it appears on the document.

For regular clients there is a subscription fee schedule. In the deal you set the day to issue the invoice, the payment term and how many days ahead to remind. From then on the platform reminds you when it is time to invoice and flags overdue payment.

The manager sees all of this on the dashboard: sales and money by payment form; how much subscription revenue is expected, invoiced, received and overdue. The “For the period” block shows revenue from closed deals, the average deal cycle, new deals and subscription fees received — for any range “from … to …”.

Sign-in security: a security log, brute-force protection and two-factor sign-in

When dozens of people use the platform, “who signed in and what did they change” stops being a theoretical question. Now the answer is on one screen.

The security log records who signed in and out and when, from which machine; failed sign-in attempts — under which login and why access was refused; changes to accounts and roles as “before → after”, with the name of whoever made them. Entries cannot be edited or removed from the screen.

Brute-force protection: after several failed attempts in a row, sign-in from that machine under that login is paused. During the pause the password is not checked at all, so guessing is pointless.

Two-factor sign-in: after the password you enter a code from a phone app (Google Authenticator or a compatible one) or a backup code. It works without the internet and without third-party services. The administrator turns it on — for everyone or only for selected roles.

And a separate export permission. Only people who have been given it can export tasks or the client base to Excel. The file goes to the browser of whoever clicked, and nothing is left on the server. Every export, and every attempt without the permission, goes into the security log.

All of this is part of the platform for every client, at no extra charge.

The “AI costs” screen: how much, on what and by whom

Language models in SensFlow are paid for with your own provider key. The bill comes from the provider, and it is hard to tell from it what exactly the money went on.

The new “AI costs” screen shows spending for the month and for today, a forecast to the end of the month, which functions the money went on and which people are spending it.

Monthly limit: at 80% you get a warning, at 100% paid AI actions stop. The rest of the platform — tasks, CRM, documents — keeps working as usual. You learn about overspending in advance, not from the bill.

Every function except the orchestrator chat can be switched off separately. And if the platform does not know a model’s price, it does not count it as zero: a task nobody is watching at that moment will stop and say so.

Email in the CRM: a client’s letter lands in their card by itself

Correspondence with a client usually sits in a sales manager’s personal mailbox. The manager goes on leave — and nobody knows what the client wrote yesterday.

Now a work mailbox connects to the CRM over the standard IMAP protocol. A letter from a known client lands in that client’s card. If the subject carries a deal tag, it goes straight into the deal card.

A letter from an unknown sender does not create a card on its own. It goes to the “Unprocessed” queue, and a person decides what to do with it. That keeps the base free of newsletters and random addresses.

One more time saver: a commercial proposal is assembled from the deal card — from its line items and the client’s details, with no retyping.

Negotiation trainer: a sales manager practises on a client played by AI

You only learn to talk to clients by talking. Practising on real clients is expensive, and role-play with a colleague quickly bores you both.

The platform now has a negotiation trainer. The manager holds a conversation with a client played by AI, and afterwards gets a review against eight criteria: scores and word-for-word quotes of their own lines. Not a vague “be more confident”, but the exact point where the conversation went wrong.

A manager with the right permission sees the reviews of the whole team and knows who needs to learn what.

The trainer is a separate module. It is switched on for the companies that need it.

Who sees what: your own cards, your unit’s cards, or all

When the whole company uses the CRM and tasks, a sales manager does not need to see the next department’s clients. A head of a division, on the other hand, needs to see everything their unit handles.

Now, on top of permissions for sections, there is card visibility — three levels: only your own (where you are the author or the person responsible), your own and your unit’s, or all. The unit comes from your org structure — no second hierarchy to build.

The level is changed in the interface, without a restart and without editing files. Lowering it really cuts access rather than just hiding a menu item: permissions are checked on the server, so someone else’s card cannot be opened even by its direct address.

Each section — CRM, Task Desk, People, Products, Files, AI costs, Security log and others — has its own permission. A role is a named set of permissions: roles are created by copying existing ones and assigned to a person or a whole unit. The platform will not let you remove the last administrator: a company cannot be left without an owner.

A financial model from a plain-language description — with live formulas

Before, the platform needed a ready-made file to build a financial model. Now it is enough to describe the business in words.

You get an Excel workbook where the numbers are calculated by live formulas, not typed in. Change a price or a volume in the parameters block, and the whole model recalculates itself. If you already have a model, the platform works with it as before.

The totals in the table are checked by a calculator, not by AI. If something does not add up, the platform tells you which cell and which two numbers disagree.

Along the way, the platform changed how it picks a system for a task. Every system now has a passport: what tasks it is for, what goes in, what comes out and what it does not do. The system is chosen by its passport, not by how similar the words in the request look.

The orchestrator looks into your cards — and proposes an action instead of taking it

You can now ask the orchestrator about a client or a deal: what stage it is at, what the deadline is, what was agreed. The card is found by code, not by the model. If several cards fit the description, the platform asks back instead of picking the “closest” one. If there is no such card, it says so.

The orchestrator can propose an action: create a task or add an entry to a card’s history. A confirmation card appears on screen saying in words exactly what will be done. A person carries it out — with a button. The entry keeps a mark that the orchestrator made it and at whose request.

What it never does: it gets no separate access to the database and cannot do more than the person who called it. It sends nothing outside, moves no money, deletes nothing, does not change a deal’s stage and signs no documents.

The platform runs as an office server: a service, autostart and an encrypted channel

SensFlow is installed on one machine in the office, and employees open it in a browser on their own computers. Now that machine behaves the way a server should.

The platform runs as an operating-system service with autostart: after the machine restarts, nothing has to be launched by hand.

Work across the office network goes over an encrypted channel. The key and certificate are issued by the operating system itself, and the certificate is added as trusted on employees’ computers. We deliberately do not offer access from outside the office — that is a product boundary, not something unfinished.

On the same day we filed the copyright registration for the platform.

An org chart nobody has to draw

We have been quiet for a while — not because there was nothing to tell, but because there was too much of it to stop and write. We start with the part that turned out to cost the most in day-to-day work.

There is now a “Company structure” screen. The platform draws the chart itself, from what is already in the system: who sits in which unit and who reports to whom. Nothing to draw separately, nothing to refresh — the chart always shows the company as it is right now.

That is deliberate. A chart maintained by hand is a chart someone forgets to update — and then decisions get made from it, and they are wrong. Here there is nothing to update: change a person or a unit and the chart is already different.

A large company still fits on one screen: branches collapse, and the platform says in words exactly who is currently hidden. Nothing is lost — it is simply folded away.

And the chart shows the gaps: someone without a manager, a unit without a head. You see it at a glance instead of after a month of things going wrong.

Company change history — a log that forgets nothing

A separate “Company change history” screen. You can see what changed and when: who was hired and who left, which units were created, moved or closed, who reports to whom. The platform records this by itself — there is no separate log to keep.

Entries do not disappear. Logs like this usually keep the last few hundred changes and quietly lose the oldest ones. Here the history grows for as long as you need — nothing is erased.

Every line says WHO made the change. And when the platform does not know that, it says so instead of attributing the change to someone at random.

History can be searched by word, filtered by period, and what you found can be sent to a colleague as a link.

Thousands of tasks — and the screen still opens just as fast

Tasks completed or cancelled more than 30 days ago are no longer read every day. So the list and the board open just as fast years into the work, when there are already thousands of tasks.

This is measured, not felt: on two thousand tasks the screen used to open in 340 milliseconds, now it takes 40. For reference: people perceive an action as instant below 100 milliseconds, and attention starts to drift somewhere past 400.

No task goes anywhere: open a card by its number and it is there as always. Need the old ones — tick “search the archive too”, and they are back.

This is the quiet kind of work you do not notice while it is there and notice immediately when it is not: the product does not get slower with the years.

Tasks inside the platform: you cannot close one without a result

The platform now has a task screen. Author, assignee, co-assignees, watchers; a board and a list; filters, your own columns, saved views, export to Excel. Every task tracker has all of that, which is exactly why it is not the point here.

The point is this: a task cannot be completed without a result. Not a checkbox in the settings that someone eventually unticks, but a condition of closing. If the work was done but left no trace anywhere, the task does not close. We do not build settings that can be switched on but not off — and we do not build gates that a checkbox can bypass.

The second thing we missed in other systems: every item of the execution plan has its own deadline. “Prepare the package by the 13th” is made of steps, and each step has its own date. The platform will not let a task be due earlier than an item inside it — and it will tell you which item is in the way.

The third is about roles. When one person is both the author and the assignee, the rights add up: no role takes away what another role granted. It sounds like a detail until it turns out that otherwise a person cannot accept their own task.

And about the AI. It reads the comments on a task and offers: if a new deadline appears in the discussion, it shows a suggestion with two buttons. The decision stays with the person, and the task feed shows who made it. The platform does not write into your task on your behalf.

The platform is assembled around you: you pay for the systems you actually use

The product used to be a single block: all of it or none. Now the make-up of your platform is your choice. Need a lawyer, a copywriter and a financial model — those are exactly what arrive. Need one more system a month later — it arrives as an ordinary update, with no reinstall.

The point here is not the price, it is the boundary. A system you did not buy is physically absent from your machine: no files, no mention in the catalogue. We do not put a lock on something we shipped — we simply do not ship it. A lock can be picked; a file that is not there cannot.

The reverse side is just as honest: if you changed nothing, your set stays exactly as it was. Silence never takes away what you already have.

Your work is yours. One-file export, regardless of payment

A question an honest platform should settle before the contract is signed: what happens to my data if we stop working together. In SensFlow the answer is simple — you take everything, at any time.

One button assembles an archive: the results of every run in ordinary formats, your correspondence with the orchestrator, your work streams, tasks, the input and output files of every system, accumulated knowledge, your client base. All of it opens without SensFlow.

Two decisions we consider a matter of principle. First: the export never asks about payment status — otherwise it would refuse at the very moment it is needed most. Second: your language-model access keys never enter the archive and stay on your machine alone.

CosmetoCard is live on Google Play — the first product the ecosystem took all the way to a store

“Cosmetologist’s cabinet” has been published on Google Play and is available in 175 countries. It is the first product the SensFlow ecosystem took through a full cycle: market research, architecture, code in .NET MAUI, an interface in four languages, a UI/UX audit, fourteen days of closed testing — and the release. Multi-agent pipelines did the work, a human made the decisions.

The app itself is about one thing: the client data stays with the practitioner. The file cabinet works fully offline — no clouds, no servers, no fees. It is free to try: 14 days and up to 5 client records, with no registration and no card required.

Why this matters beyond beauticians: we walked the road from an empty folder to a store listing using the very method we sell. That is our answer to “do you use this yourselves”.

Tender counsel: from a tender number to deadlines, rights and an action paper

The hardest part of public procurement is not writing the bid — it is not missing a deadline. Different procedures carry different time limits, and a person running ten tenders keeps those days in their head or in a notebook. Now the system keeps them.

You give a tender number — SensFlow pulls its record from Prozorro, identifies the procedure type, retrieves the norms that apply to that exact type, and returns a table of “norm → deadline → in your tender”, with steps and the legal basis for each. Alongside it comes something nobody usually writes down: which rights do NOT apply to you, because your procedure simply does not grant them.

The norm base is verified against the official text on the Verkhovna Rada portal — verbatim quote, article, date of verification. Eight procurement regimes are covered. And most importantly: with no tender number, the system refuses to name any days instead of guessing. A deadline borrowed from someone else's procedure costs more than a missing answer.

Next to it runs tender discovery: an in-house index built on the free Prozorro API plus a profile of your products. The qualifier ranks what it finds by prospect, but discards nothing — whether your product fits remains your call.

The deadline document is assembled without a language model — code writes it from the verified base. So it contains no guesses and no “almost right” numbers. Self-declared “verified” and “PASS” claims are stripped: the verdict is set by code on the fact of the check, not by a model about itself.

SensFlow gets a face of its own: brand identity and logo

The platform now has a recognizable look of its own. SensFlow now has a brand mark and a unified style that runs through everything: the website, the work console and the Telegram bots. The product that works for you every day now looks like a coherent brand, too.

And it is not only about looks. The documents you receive now come out properly styled: Word on a branded letterhead, presentations with the logo, reports with a branded header. A finished, professional look you can confidently put in front of management or partners.

The orchestrator has become smarter and more honest: a complex task is led by a team, not a lone agent

We have brought an important principle back to the orchestrator. A simple task it does itself. But when a task is complex and there is no ready system for it, it no longer answers off the cuff — it offers to assemble a full team of AI agents for it through the built-in builder.

And if you do still ask for a quick one-off answer — the platform honestly marks it: 'a quick one-off result without full verification'. No hidden compromises: you always know when a verified team was at work and when it was a quick answer.

The site now has a “Built products by the Orchestrator” showcase. First product: Beautician’s Cabinet

SensFlow is more than a platform — it’s a production line. The site now has a Built products section — a showcase of products that went through a full cycle inside the multi-agent ecosystem: from market research and architecture to code, testing and release.

The first exhibit is Beautician’s Cabinet (CosmetoCard): a private client file cabinet for beauticians on an Android tablet, smartphone and Windows. Fully offline — the data stays with the practitioner. The Android version is heading to Google Play, and we are already recruiting testers.

Update delivery passed its baptism by fire: one button — and the client has the latest version

The SensFlow update mechanism completed its full loop on a real client’s machine. We published an update — the client saw it in their console, clicked “Update” and had the latest version of the platform a minute later. Data, chats and saved keys stayed in place.

Under the hood it’s done properly: before applying, the platform makes a backup on its own, every update is signed and integrity-checked, and changes can be rolled back if needed. This is how we support every license: the client just works — the platform keeps itself up to date.

The interface is now multilingual — and runs on Windows and Mac

SensFlow working console now speaks three languages. The switch is right in the interface: the language applies instantly and is remembered, and each build can start in the language you need out of the box.

At the same time the platform is now packaged for macOS, not only Windows: one file, a double click — and the console opens. This makes SensFlow available to international clients and partners regardless of their OS and language.

AI technical-support bot: works on its own, reads screenshots, hands off to a human

SensFlow's AI technical-support bot is now live. It was born from a real case — while deploying the platform on a client's machine, when urgent help from an IT specialist was needed. The bot answers clients on its own, reads the screenshots they send, speaks in a tone matched to the client's profile, and hands off hard cases to a human right away. The principle is simple: the user comes first.

Stealth mode: full local isolation for trade secrets

Stealth mode has arrived — for clients with strict confidentiality requirements. Switch on a local model, and the platform analyzes your documents right on your machine: sensitive data never leaves the perimeter.

Ideal for contracts, financial reports and any materials containing trade secrets. You choose the level of privacy yourself: fast cloud for everyday tasks, or full local isolation where the cost of a leak is high.

The site now speaks English — for international clients

The entire SensFlow site is now available in English — just click the UA | EN toggle in the menu. This opens the platform to international clients and cross-border projects.

We also polished the mobile version: navigation and all sections now open conveniently on a smartphone.

The Legal Assistant now works across two legal systems

SensFlow’s legal system has reached a new level: it now relies not only on Ukrainian law but also on Kazakhstan’s. This makes it possible to prepare and analyze cross-border agreements and legal documents, and to support deals across two legal systems at once.

A methodology has been developed for adding other countries’ legislation and branches of law. An independent quality check was added that verifies every conclusion against current norms. It’s a truly powerful tool for a corporate lawyer.

Teamwork: roles and operator seats

SensFlow is now ready for a whole team. New roles have appeared: an administrator with full access and operators with limited rights — for safely scaling the platform to your employees’ workstations.

Each operator works with the systems within their role, while management and settings stay with the administrator. Convenient and safe for the whole team.

Business Analyst: enhanced with Eric Partaker’s operational frameworks

The Business Analyst system has been significantly enhanced: its strategy and analytics agents were extended with the operational frameworks of Eric Partaker — a leading CEO coach (ex-McKinsey and Skype, author of the bestseller “The 3 Alarms,” “CEO of the Year”).

This noticeably widened the system’s product range — from deep startup work to KPIs, in-depth analytics and preparing CEO-level management decisions.

Presentations and dashboards — now with real charts

The platform has learned to deliver ready-made visual materials: interactive HTML dashboards and presentations with real charts, not just tables.

One file — open it and show a client or management. Analytics became not just accurate, but visual.

Data you can trust: every figure from verified sources

We strengthened fact control: for analytical tasks, the platform checks against real sources online and doesn’t present unverified figures as fact.

Less “made up” — more trust in the result, especially where the cost of error is high.