You see what the AI proposed
Every proposal lands in a shared registry instead of getting lost in one person’s chat. The whole team looks at the same picture and sees the same history of changes.
Platform · Process engine BPMN runtime
A BPMN diagram here is not a picture for a slide deck. AI drafts the diagram, the forms and the triggers; you press Publish and the process is live a minute later.
A diagram stops being a picture for a slide deck and becomes the thing that actually runs.
Direct message
Platform AI is a full runtime for business processes. You connect through the MCP server and describe the process you need in plain language. The platform assembles a working BPMN diagram with roles, forms and triggers and puts it in the manager’s approval queue — once it is confirmed, the process goes to production and starts doing the work. Not a chatbot that tells you what it would do, but a colleague that does it.
Processes are assembled from pre-built, tested blocks — model, roles, BPMN, forms. You do not generate structure from scratch, so it takes an order of magnitude fewer tokens.
Every node becomes a real action: a task in the registry, an email, an API call, a status change. Nothing sits between describing a process and running it.
Every process and diagram change lands in the manager’s approval queue — where the rest of the approvals live, confirmed with one button.
Everything the agent proposes arrives in one queue with a diff and a version history. You approve it the way you would a pull request — without the code and without programmers on the team.
Every proposal lands in a shared registry instead of getting lost in one person’s chat. The whole team looks at the same picture and sees the same history of changes.
Every change comes with a diagram diff, a log of previous versions and the option to reject with a comment. Any step can be postponed, accepted or sent back with a remark.
“Make the escalation 12 hours instead of 24” is enough — the agent updates the BPMN itself and saves a new version. You do not have to learn the notation or call a developer.
Processes start on events and on schedule, SLA timers and escalations are built in. Nothing waits for someone to press a button in the morning.



The platform connects to what the company already runs on: ERP, CRM, corporate email, messengers, document storage and any external service over REST or SOAP. Data and communication channels meet in one place, so a process does not have to call ten systems one by one.
Directories of employees and departments, registries of requests and cases, forms of any complexity, role-based access and the permission matrix, the status of every process instance. AI agents read and write by the same rules as employees, so the security perimeter and the audit trail stay intact.
Processes start as a reaction to events — a record created, an incoming email, an expired SLA, a fired escalation — or simply on a schedule. Steps inside are carried out by people and AI agents; time control and escalations are built into the platform.
Count what you stop spending
Approvals, hand-offs and reminders that used to move by hand run as a published process. And because every process is assembled from pre-built, tested blocks instead of generated from scratch, an AI agent needs roughly ten times fewer tokens to build one.
Uses
Customer stories
Not “everything got better after the rollout”. A specific scenario, a measurable metric before and after, and a specific person who now uses it every day.
1 controller
instead of 4 on shifton the first line
![]()
First line used to be four pairs of eyes and a pile of chats. Now the agent closes seven requests out of ten on its own, and an engineer only ever sees the hard ones.
Service desk · B2B SaaS
Four operators on shift, a pile of chats and no way to tell which request was already answered last week. Every recurring question was solved again from zero.
Requests from the portal, email and messengers land in one registry with statuses, owners and deadlines.
Three statuses, a 24-hour SLA timer and an escalation — drafted by the agent, approved by the head of support.
Recurring questions are closed from decisions already taken; anything new goes to an engineer with the context attached.
Every change to the process arrives in the approval queue with a diff. Nothing goes live on the agent’s word.
First line used to be four pairs of eyes and a pile of chats. Now the agent closes seven requests out of ten on its own, and an engineer only ever sees the hard ones.
4 hours
instead of 3 working daysto approve an invoice
![]()
The invoice used to travel by email and stall on whoever was away. Now the timer escalates it on its own, and I see where it is standing without asking anyone.
Finance · Manufacturing group
The invoice travelled by email and stalled on whoever was away. Nobody could say where it was standing without asking three people.
Three statuses, an owner at each step, and the document attached to the record rather than to a thread.
After 24 hours the invoice escalates to the curator by itself — no reminder emails from a person.
ERP and the document storage are on the same platform, so the amount and the counterparty are not retyped.
The finance director sees where every invoice is standing, without asking anyone.
The invoice used to travel by email and stall on whoever was away. Now the timer escalates it on its own, and I see where it is standing without asking anyone.
1 day
instead of 5to onboard a new hire
![]()
Accounts, access, equipment and the security briefing used to be five separate chats. Now it is one process, and on day one the person already has everything.
HR · Retail chain
Accounts, access, equipment and the security briefing lived in separate conversations. On day one the new hire usually had none of them.
From the offer to the first working day, with the owner of every step named.
The process fires when the record appears in the employee registry — nobody has to remember to launch it.
IT, security and the office work at the same time instead of one after another.
Who did what and when stays in the run log, so a missed step is visible before day one.
Accounts, access, equipment and the security briefing used to be five separate chats. Now it is one process, and on day one the person already has everything.