LEAN TOOLS · POWERED BY LNSE

LEAN TOOLS Say what your plant does. Get the map, the cell, the year and the kaizen you can defend in a review.

Value stream mapping is the most useful thing lean gives you and the most tedious thing to do by hand — a day of walking the floor, then a week of drawing boxes, and a future state that comes down to whoever argues hardest. Describe the stream in your own words instead — or pick your industry from the list and have a whole worked stream drawn in one click. LNSE draws the current state from what you said, tells you which numbers it had to assume, and answers the eight questions of Learning to See. Then it does the three things a map on its own never did: it proposes the production cell each future state can actually run, it levels the year so February is not run at a different pace to hit March's volume, and it schedules the kaizen that gets you there. Deterministic, auditable, free to use, and no sign-up.

Free No account, no card, no per-map charge
1,201 Worked example streams to start from, thirty in every industry
99.4% Of 16,695 planted defects fully caught on the 20,000-map release gate
5 Future states, each with its cell, its levelled year and its kaizen Gantt
Open the VSM Generator — Free See how it works
THE LEAN TOOLBOX

One Engine. A Growing Set of Tools.

Everything here runs on LNSE, so every tool reads the same substrate and gives the same answer twice.

account_tree FREE · LIVE NOW

Value Stream Map Generator

Speak or type the story of one product family — what you make, who takes it, what it goes through, how long each step takes, and how much stock sits between them. You get the current state drawn to the standard iconography, the eight questions answered, five future states ranked with the kaizen each needs, a cell layout and a levelled year for every one of them, every KPI the catalog can derive, and a printable document with the whole analysis in it — maps, cells, heijunka and Gantt, nothing clipped.

  • record_voice_overDescribe it in plain words, spoken or typed — no forms, no seeds, no lean vocabulary required.
  • playlist_add_checkOr start from your industry: 1,201 complete streams — at least thirty per industry — raw material to customer, one click to a map you can then edit.
  • help_outlineIt asks when it needs to, and marks what it assumed, so nothing quietly invents itself onto your map.
  • alt_routeFive future states — flow first, pull first, level and pace — each with the cell it can run and the year it levels to.
  • printPrint one future state or all five. The cell, the heijunka and the kaizen Gantt travel with the map they belong to.
  • forumSeveral streams at once: one chat per value stream, switch between them as you go.
Open the tool arrow_forward
construction IN DEVELOPMENT

The Rest of the Toolbox

Heijunka, the cell and the kaizen schedule now ship with the map. What is still being built on the same substrate is the paperwork a plant actually hangs on the wall after the review.

  • scheduleValue stream plans with owners and review dates, generated from the kaizen the map already named.
  • descriptionA3s and standard work sheets that inherit the map's own numbers instead of restating them.
  • precision_manufacturingShop-floor andons and short-cycle control wired to the same pacemaker the year is levelled on.
Tell us which you need first arrow_forward
NEW ON THE GENERATOR

A Future State Is Not a Picture. It Is a Cell, a Year and a Schedule.

Three drawings now travel with every one of the five strategies — on screen, and on the print.

view_in_ar

The production cell it can actually run

U-shaped, straight line, or isolated islands — chosen from the flow and cell-design rules the engine already holds. Monuments stay outside the cell, with a reason. Manning, takt, operator loops and standard WIP are on the drawing, not in a footnote.

calendar_month

A year levelled, not a monthly target

Heijunka takes the future-state map and builds the schedule it implies: one daily rate for the year, monthly volume that follows that month's working days, a pitch pattern — the heijunka box — and staging posts so the year contains maps, not just volumes. February is no longer run at a different pace to hit an equal-monthly number.

view_week

The kaizen, on a Gantt, for that state

Each future state keeps the bursts that get you there, sequenced and dated. Print one state and you get its Gantt. Print the full analysis and you get all five — so a reader never compares the wrong schedule with the wrong map.

print

Nothing cut, on screen or on paper

Station names, burst captions, pitch figures and checks are drawn whole. The panes scroll on both axes; the print sheet scales the drawing to A3 instead of clipping it. A number ending in an ellipsis is a different number, and the generator refuses to print one.

START FROM YOUR INDUSTRY

1,201 Streams Already Written. Pick Yours and Press Generate.

The hardest map to draw is the first one, so there is one waiting for almost every plant and office we could think of.

factory

Thirty a trade, 38 trades, 972 departments

Foundry and forge, machining, stamping, pharmaceutical, food, print, footwear, shipbuilding, jewellery — and the transactional side too: claims, lending, service desks, outpatient clinics, admissions, distribution centres, restaurant kitchens. Every trade has at least thirty streams, so the picker offers a brake disc, a manifold and a pump casing rather than one generic casting.

linear_scale

Whole streams, not fragments

Each one runs raw material to customer, with the demand, the working calendar, every operation's cycle time, changeover, uptime and scrap, and the stock sitting between them — everything a map needs, so nothing is asked of you. The arithmetic is the trade's own: 120,000 blanks a month on a two-second takt in one, a third of a boat a day in another.

bolt

One click, then yours to change

Two dropdowns and Generate. Or copy the narrative into the chat and edit it into your own plant — change the demand, rename the stations, move the stock, and redraw.

verified

Every one of them commits

All 1,201 pass the verifier without a single clarifying question, and are checked on every build. An example that made the engine ask you something would be an example that failed at the one job it has.

They are worked examples rather than templates to fill in: the numbers in each one are internally consistent, so the takt, the loading and the lead time you see are the arithmetic of that stream and not a placeholder. Find the one nearest your own shop, read what it says, and edit from there.

FROM A SENTENCE TO A MAP

You Describe It. It Draws It. Then It Argues Its Case.

Four steps, and the only one that needs you is the first.

01 mic

Say it the way you'd tell a colleague

“We stamp steel brackets for State Street Assembly, 18,400 a month, two shifts…” — typed or spoken. It recognises the companies, the operations, the cycle times, the stock and the schedule out of ordinary prose. Or pick your industry and start from a stream that is already written.

02 draw

The current state, drawn from your words

Standard icons, real data boxes, the stock triangles where you said the piles are, and a lead-time ladder that puts your processing time against your lead time. Every figure traces back to the sentence it came from.

03 quiz

It asks — and admits what it assumed

Only what it genuinely needs, one question at a time, in plain language. Anything it had to supply from the substrate is listed as an assumption, with the reason, so you check the right numbers first.

04 insights

Future states you can defend — and run

Takt, flow, pull, the pacemaker, levelling and the required process improvements — the eight questions, answered in writing. Five strategies, each with its cell, its levelled year, its kaizen Gantt and what it costs you.

LNSE

The Lean Neuro-Symbolic Engine. Not an LLM.

Same architecture and the same mathematics as MNSE on the medical side — pointed at manufacturing.

menu_book

A substrate, not a prompt

686 versioned substrates in 14 families — processes, metrics, mechanisms, waste, rules — with 47,138 phrase aliases. If a concept is not in the substrate, the engine will not draw it.

rule

Rules that fire, not text that sounds right

Every finding on a map is a named rule with a category — convention, discipline, design, improvement. You can argue with a rule. You cannot argue with a paragraph.

fact_check

A verifier that refuses

A map is committed only when the verifier clears it against a calibrated threshold. When it will not, the tool says which finding blocked it rather than shipping something that reads well.

bolt

CPU, milliseconds, $0 a call

No GPU, no tokens, no per-call bill. Which is exactly why the tool can be free: drawing your map costs us electricity, not an API invoice.

0.904 Encoder macro-F1 on a 1,681-span holdout (v31)
95.6% Verifier precision at the commit threshold (v31)
0.953 Verifier AUC on a 4,393-map holdout
100% Of 3,305 clean maps on the gate, pass their own rules

Models v31, substrate 1.2.0, certified 16 August 2026. The training corpus is 100,000 adversarial maps with deliberate faults — a pacemaker in the wrong place, pull without kanban discipline, a takt computed on gross time. The release gate then replays 20,000 maps against the promoted models: 99.4% of 16,695 planted defects fully detected, zero commits with a visible defect on the page, and every future state improves lead time (200 / 200). A model with a beautiful AUC that commits one corrupted map has failed; this gate is about that, not about a loss curve.

Reading is drilled the same way, and against the harder standard. A second corpus writes 400,000 narratives about the products in that example library, across all 38 industries, and says every fact in a different way each time — the demand monthly, weekly, daily or annually; the stations numbered, lettered, dashed or unnumbered; scrap as a reject rate, a loss, a percentage or “none at all” — and then checks the engine read back exactly what was said. It reads every one of the 29,018,255 facts those narratives state: nothing misread, nothing missed, and nothing invented that was never there. Three quarters of them are a holdout the engine was never tuned against, and it scores the same on those as on the rest.

NO CATCH

Free, Because It Costs Us Almost Nothing.

A neuro-symbolic engine has no token bill. It runs on a couple of CPU cores and answers in milliseconds, so there is no per-map cost for us to pass on to you. Use it for your plant, your client's plant, or your students — no account, no card, no watermark on the print-out.

Map a Stream in the Next Ten Minutes.

Bring one product family and the numbers you already know. The engine will ask for the rest and tell you what it had to assume.

Open the VSM Generator License LNSE for your plant
✓

Message sent successfully!