// the problem

Operations run on knowledge that was never written down

In most non-digital-native businesses, the real process lives in one person's head and in muscle memory on the warehouse floor. How an order actually gets reconciled, which exception gets escalated and which gets absorbed, the unspoken sequence everyone just knows.

When that person is out, the process is out with them. Writing it down by hand is slow, goes stale the moment anything changes, and nobody reads a 40-page document. So the knowledge stays trapped, and nothing downstream can be automated because nothing upstream is described.

// how it solves it

Watch the process once. Get living process maps

FlowTwin observes a process a single time, a screen recording or a walkthrough, and turns it into structured process maps and runbooks. It captures the steps, the decision points, and the exceptions, not just the happy path.

Everything it produces is editable and adjustable, so a human stays in control, and auditable, so every step traces back to what actually happened in the recording. Because the output is generated rather than typed by hand, it stays cheap to regenerate when the process changes, which means it stays current instead of rotting in a shared drive.

// architecture overview

Capture, segment, structure, publish

01
CaptureScreen or video recording of the process as it really runs
02
SegmentVision model (SAM) splits the run into discrete steps and actions
03
StructureAn agent extracts steps, decisions and exceptions into a typed schema
04
PublishEditable runbooks, process maps and diagrams, adjustable and auditable

OODAA loop wraps the pipeline: each run is observed, oriented, and assessed so quality improves over time.

// what makes it different

Not a static document. A system that keeps itself honest

self-learning
Self-learning loop

An OODAA loop wraps every run. FlowTwin observes its own output, orients on what it got wrong, and improves extraction quality over time, without being hand-tuned for each new process.

auditor
Built-in auditor

An auditor agent checks each generated map against the source recording, flags missing steps, ambiguous decisions and gaps, and refuses to publish a process it cannot fully account for.

editable
Editable & adjustable

Maps and runbooks are not locked artifacts. Edit a step, reorder a branch, add an exception, and the structure stays valid and exportable for humans and machines alike.

auditable
Fully auditable

Every step links back to the moment in the recording it came from, and every change is versioned, so you can prove how a process was defined and when it changed.

// features & updates pipeline

Where FlowTwin is heading

plannedFlowTwin Explorer

An analyzer engine on top of captured processes. It reads a process and proposes concrete automation and agentic opportunities, scoring where an agent could take over and where a human still has to stay in the loop.

plannedSelf-hosted capture app

A user-run desktop app that helps a person capture the processes they do across a normal working day, privately and on their own machine, with no recordings leaving their control.

plannedPhysical capture module

A video capture module for physical work on the factory floor or in the warehouse, extending FlowTwin from screen-based processes to the hands-on operations that never touch a keyboard.