● B2B offer — Industry & production quality
Production monitoring: drift seen before the scrap
Process drift shows up in the readings before it shows on the parts. Your agent tracks the data from your sensors and your checks, compares each reading with the ranges you have set and alerts as soon as a trend moves away, with the measurements that document it. Hosted in France: your process data and your machine settings stay with you. The production manager decides on any intervention.
Hosted in France
Process data protected
GDPR & AI Act: governed deployment
Human oversight
Updated on
Deployed in a few weeks
Production monitoring · hosted in France
Where do the parameters on line 2 stand?
Readings within the ranges set across all the check points, bar one.
One trend has been moving away from the target value since the start of the shift: the successive measurements are provided.
The range set and the deviation observed are shown side by side.
🔗 Sourced · sensor readings and the ranges set
One trend has been moving away from the target value since the start of the shift: the successive measurements are provided.
The range set and the deviation observed are shown side by side.
🔗 Sourced · sensor readings and the ranges set
Should we stop the line?
I provide the measurements, the trend and the applicable range so the decision can be taken quickly.
Stopping a line, adjusting a machine or setting a batch aside commit production and safety: those decisions belong to the manager.
✎ Support · measurements provided, operations decision
Stopping a line, adjusting a machine or setting a batch aside commit production and safety: those decisions belong to the manager.
✎ Support · measurements provided, operations decision
Local inference · no data outside the EU
Readings hosted in France
Sovereign by designLocal inference or hosting in France
GDPR & AI Act: governed deploymentTraceability & human oversight
TurnkeyDesigned, installed and operated for you
The production manager decidesThe agent prepares, never rules
✦ In brief
A Blue Lemon Agent monitoring agent tracks your sensor readings and checks, compares them with the ranges you have set and alerts as soon as a trend moves away, backed by the successive measurements. Stopping a line or adjusting a machine remains a production decision. It runs on local inference or is hosted in France: your process data stays with you, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
100%
hosted in France in the target architecture
0
transfer outside the EU in the target architecture
6
monitoring uses ready to deploy
0
decision taken without human approval
These figures describe our offer, not results measured at a client. How large the gain is on your number of lines and check points tracked is confirmed by a pilot.
The context
What does an AI agent bring to your production monitoring?
A trend spotted in the readings leaves time to adjust; found on the parts, it leaves a batch to deal with.
! The issue
Quality in production shows in the trend of the readings as much as in their instantaneous value. Following that trend across every check point, shift after shift, calls for continuous attention. The agent keeps it up and documents every signal with the successive measurements and the applicable range.
✓ Our answer
The production manager has documented signals, across every check point and not just the most closely watched. Stopping a line, adjusting a machine or setting a batch aside commit production, safety and sometimes the product's conformity: those decisions stay human. Local inference or an isolated resource hosted in France: your process data and your machine settings do not leave the company.
The decisive point
Your process data and your machine settings: sovereignty & compliance
Your process data and your machine settings are industrial know-how. Here is how the architecture of our agents protects them.
Local inference
The agent can run on a machine belonging to your organisation: no process data and no machine setting leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your production lines and your readings: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For your process data and your machine settings, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your industrial site and its control ranges.
Every alert tied to its measurements
Successive measurements, the range set and the deviation observed are kept for every alert; encryption, role-based access and logging that can be used in a quality audit.
AI Act: governed deployment
The agent is strictly in support; no line is stopped, no adjustment is applied and no batch is set aside automatically; traceability and human oversight from end to end.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
For processes covered by a patent, an approval or a manufacturing secret, SecNumCloud and reinforced hosting are options depending on your requirements. A single architecture is designed to answer both the GDPR and extraterritorial exposure. Designed for deployment in line with the GDPR and the AI Act, after the processing, roles and context-specific risks have been assessed.
Demonstration
See the agent at work
5 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
The company in this demonstration
Fictional companyPolvane — technical plastic injection moulding, one production site
- Sector
- Technical plastic injection moulding — parts for the automotive and appliance industries
- Headcount
- 210 staff, including 12 quality technicians and 6 setters
- Customers served
- 34 European industrial customers, including 3 car manufacturers
- Volume
- 4 injection presses, 128 control points, 2.4 million parts a month — around 1,250 parts an hour per press
- Tools in place
- Shop-floor supervision, production MES, maintenance CMMS, control laboratory and three binders of manual readings — the agent plugs into them, nothing is replaced
- Who decides
- The production manager signs every setting change and every line stop; the quality manager rules on the fate of a batch
- Room for improvement
- Trend watching takes 60 % of a quality technician's time and covers only 22 of the 128 points; scrap runs at 1.9 %, or €86,600 a month; a drift is spotted 4 h 20 after its first sign on average
Polvane runs three shifts, seven days a week on two of its four presses. Its readings already exist — sensors, laboratory, binders — but nobody has time to read them all. The agent runs on local inference on a machine at the plant and plugs into the shop-floor supervision, the MES, the CMMS and the laboratory: it reads every point, compares it with the ranges written by the quality department and documents each signal with its readings. The exchanges below cover one quarter, from the reopening of the history to the review.
This company, its figures and the exchanges that follow were invented for the demonstration. They illustrate a common situation; they describe no real client.
Production supervision · hosted in France
I read your last 18 months of readings overnight — 41 million measurements across 128 control points — and found your 41 real drifts in them. Thirty-five of those announced themselves in the figures well before they showed on the parts.
A drift is a run of measurements moving steadily away from the target while staying, at first, inside the permitted range. That moment is exactly the one you can recover with a setting change, without scrapping anything.
The gap I measured, and it is the gap that decides the gain: your drifts were spotted 4 h 20 after their first sign on average. At 1,250 parts an hour, that is 5,400 parts already made by the time the decision is taken.
What the same history gives with continuous reading: the 35 announced drifts flagged themselves after 11 minutes on average, or 230 parts. Twenty-three times fewer parts involved, on exactly the same readings.
And here is where it plays out: 29 of the 41 drifts appeared on the 106 points nobody had time to follow, not on the 22 your technicians watch closely. It is not your sensitive points that cost you, it is the others.
What I suggest: that your production manager pick three past drifts he knows by heart, and that I show him what my readings would have displayed, hour by hour. You judge on cases you lived through, not on a promise. 18-months-of-readings_41-drifts-replayed.pdf4 h 20 → 11 minutes, 5,400 parts → 230
⛓ Sourced · shop-floor supervision, MES, control laboratory — 18 months, 41 million readings, 41 confirmed drifts
A drift is a run of measurements moving steadily away from the target while staying, at first, inside the permitted range. That moment is exactly the one you can recover with a setting change, without scrapping anything.
The gap I measured, and it is the gap that decides the gain: your drifts were spotted 4 h 20 after their first sign on average. At 1,250 parts an hour, that is 5,400 parts already made by the time the decision is taken.
What the same history gives with continuous reading: the 35 announced drifts flagged themselves after 11 minutes on average, or 230 parts. Twenty-three times fewer parts involved, on exactly the same readings.
And here is where it plays out: 29 of the 41 drifts appeared on the 106 points nobody had time to follow, not on the 22 your technicians watch closely. It is not your sensitive points that cost you, it is the others.
What I suggest: that your production manager pick three past drifts he knows by heart, and that I show him what my readings would have displayed, hour by hour. You judge on cases you lived through, not on a promise. 18-months-of-readings_41-drifts-replayed.pdf4 h 20 → 11 minutes, 5,400 parts → 230
⛓ Sourced · shop-floor supervision, MES, control laboratory — 18 months, 41 million readings, 41 confirmed drifts
What exactly do you track? We don't have the same sensors everywhere, and some of it is still on paper.
All 128 points, with no exception — including the 41 that existed only in your binders. They have been in since the 6th, and they have already brought in two drifts.
Where each measurement comes from:
· 74 points from the shop-floor supervision — temperatures, pressures, cycle times, clamping force. Read every 90 seconds.
· 13 points from the laboratory — dimensions, mass, mechanical tests. One to four times a shift according to the control plan — the document that states which point is measured, how often and by whom.
· 41 points read by hand — material moisture, appearance, visual inspection. Entered on the shift terminal instead of the binder, and the technician spends 4 minutes a shift on it instead of 15.
What taking the binders on immediately returned: material moisture on press 2 had been recorded for eleven months without ever being compared with anything. Replayed across those eleven months, it explains 2 of the 41 drifts — the ones whose cause had stayed “undetermined” in your file.
What you do not change: your sensors, your controllers, your MES and your CMMS stay in place. I read, I replace nothing, and no control point is added to your control plan unless you have written it.
What I suggest: that the 41 manual points move to the shift terminal on all four presses. Eleven minutes returned per shift per technician, or 46 hours a month across your three shifts — and above all, a measurement that becomes comparable instead of sleeping in a binder. 128-control-points_sources-and-frequencies.pdf74 from controllers, 13 laboratory, 41 taken from binders
⛓ Sourced · shop-floor supervision, laboratory, manual reading binders taken on across 11 months
Where each measurement comes from:
· 74 points from the shop-floor supervision — temperatures, pressures, cycle times, clamping force. Read every 90 seconds.
· 13 points from the laboratory — dimensions, mass, mechanical tests. One to four times a shift according to the control plan — the document that states which point is measured, how often and by whom.
· 41 points read by hand — material moisture, appearance, visual inspection. Entered on the shift terminal instead of the binder, and the technician spends 4 minutes a shift on it instead of 15.
What taking the binders on immediately returned: material moisture on press 2 had been recorded for eleven months without ever being compared with anything. Replayed across those eleven months, it explains 2 of the 41 drifts — the ones whose cause had stayed “undetermined” in your file.
What you do not change: your sensors, your controllers, your MES and your CMMS stay in place. I read, I replace nothing, and no control point is added to your control plan unless you have written it.
What I suggest: that the 41 manual points move to the shift terminal on all four presses. Eleven minutes returned per shift per technician, or 46 hours a month across your three shifts — and above all, a measurement that becomes comparable instead of sleeping in a binder. 128-control-points_sources-and-frequencies.pdf74 from controllers, 13 laboratory, 41 taken from binders
⛓ Sourced · shop-floor supervision, laboratory, manual reading binders taken on across 11 months
In concrete terms, what is that worth to us?
On your own figures: €86,600 of scrap a month, of which €53,700 comes from slow drifts — the ones visible in the readings before they are visible on the parts.
Scrap is parts that are made and then thrown away because they fail inspection. Your fully loaded cost is €1.90 a part, materials, machine and labour included.
The calculation, line by line, and it is yours:
· 2.4 million parts a month, 1.9 % scrap: 45,600 parts.
· At €1.90 a part: €86,600 a month, €1.04 m a year.
· 62 % of those parts come from a slow drift, according to the reason written on your own scrap tickets: €53,700 a month.
What 11 minutes instead of 4 h 20 shifts on that share: each drift caught early involves 230 parts instead of 5,400. Across the 27 slow drifts of a year, that is the difference between 146,000 parts exposed and 6,200.
And the gain does not stop at the scrap bin: a drift caught at 11 minutes is corrected by a setting change on a running press. Caught at 4 h 20, it has already triggered a batch sort — and your scrap tickets carry 11 hours of sorting a month, done by operators who should be producing.
What I suggest: start with the two presses that run seven days a week, because they are the ones producing during the hours when nobody reads the readings. Approve the scope and I hand you the first trend statement tomorrow morning. cost-of-scrap_what-slow-drifts-weigh.pdf€86,600/month, of which €53,700 from slow drifts
⛓ Sourced · scrap tickets, MES, fully loaded part cost supplied by management accounting
Scrap is parts that are made and then thrown away because they fail inspection. Your fully loaded cost is €1.90 a part, materials, machine and labour included.
The calculation, line by line, and it is yours:
· 2.4 million parts a month, 1.9 % scrap: 45,600 parts.
· At €1.90 a part: €86,600 a month, €1.04 m a year.
· 62 % of those parts come from a slow drift, according to the reason written on your own scrap tickets: €53,700 a month.
What 11 minutes instead of 4 h 20 shifts on that share: each drift caught early involves 230 parts instead of 5,400. Across the 27 slow drifts of a year, that is the difference between 146,000 parts exposed and 6,200.
And the gain does not stop at the scrap bin: a drift caught at 11 minutes is corrected by a setting change on a running press. Caught at 4 h 20, it has already triggered a batch sort — and your scrap tickets carry 11 hours of sorting a month, done by operators who should be producing.
What I suggest: start with the two presses that run seven days a week, because they are the ones producing during the hours when nobody reads the readings. Approve the scope and I hand you the first trend statement tomorrow morning. cost-of-scrap_what-slow-drifts-weigh.pdf€86,600/month, of which €53,700 from slow drifts
⛓ Sourced · scrap tickets, MES, fully loaded part cost supplied by management accounting
The ranges were written by our quality department. Are you changing them?
I apply yours exactly as written, and I display them with every alert — the range, the date it was written and the person who signed it. And I replayed them across 18 months, to tell you what they produce.
A range is the interval within which a measurement is judged conforming: a lower bound and an upper bound, set by your quality department control point by control point.
What the review shows: 96 of your 128 ranges were written when the presses were commissioned, in 2019, and have not been replayed since. They are correct; they simply judge a single measurement, and a drift is never a single measurement.
The exact count, over 18 months: 3,100 out-of-range alerts, for 41 real drifts. One useful alert in seventy-six — which is why your technicians ended up watching only 22 points.
What I do on top, and what nobody has time to do: I have written nine trend rules and run them across your 18 months of readings. A trend rule judges a run of measurements rather than a single one: it looks at the direction and the speed of the movement. For each one I give you the sentence in plain words, the number of signals it would have produced, the share confirmed at the next inspection, and the parts it would have saved.
The signature stays with the production manager: a rule enters service only once signed, and that is what makes it stand up in an audit. I save you the writing and the measuring; the decision takes ten minutes, on figures. 128-ranges-replayed_over-18-months.pdf3,100 alerts for 41 real drifts
⛓ Sourced · quality department range register, 18 months of readings, 41 confirmed drifts
A range is the interval within which a measurement is judged conforming: a lower bound and an upper bound, set by your quality department control point by control point.
What the review shows: 96 of your 128 ranges were written when the presses were commissioned, in 2019, and have not been replayed since. They are correct; they simply judge a single measurement, and a drift is never a single measurement.
The exact count, over 18 months: 3,100 out-of-range alerts, for 41 real drifts. One useful alert in seventy-six — which is why your technicians ended up watching only 22 points.
What I do on top, and what nobody has time to do: I have written nine trend rules and run them across your 18 months of readings. A trend rule judges a run of measurements rather than a single one: it looks at the direction and the speed of the movement. For each one I give you the sentence in plain words, the number of signals it would have produced, the share confirmed at the next inspection, and the parts it would have saved.
The signature stays with the production manager: a rule enters service only once signed, and that is what makes it stand up in an audit. I save you the writing and the measuring; the decision takes ten minutes, on figures. 128-ranges-replayed_over-18-months.pdf3,100 alerts for 41 real drifts
⛓ Sourced · quality department range register, 18 months of readings, 41 confirmed drifts
Nine more rules and it will be ringing all day. We already stopped listening.
It is the opposite: 3,100 alerts become 640, and 512 of them are confirmed at the next inspection. You go from 1.3 % useful signals to 80 %.
Three of the nine rules, with their measurement:
· Rule 3 — five consecutive holding-pressure readings more than 1.5 bar below target. 41 signals over 18 months, 34 followed by scrap within two hours, 7 with no sequel. Those 34 represent 18,400 parts.
· Rule 5 — melt temperature rising by more than 0.3 °C per reading over seven readings. 26 signals, 21 confirmed, all on presses 3 and 4.
· Rule 8 — cycle time lengthening by more than 0.2 seconds within an hour. 19 signals, 17 confirmed — and it is the only one that announces wear rather than a setting.
And the range that was waking you for nothing: material moisture on press 2 produced 620 alerts for 4 real drifts. I have written its tightened version and run it across the 18 months: 71 alerts, and all 4 drifts still caught. 549 fewer shop-floor call-outs, not one drift lost. Sign it and it is in service at the next start-up.
What that changes for your technicians: an alert confirmed four times out of five gets read. That is the condition for the 128 points to be genuinely watched, and not merely connected.
What I suggest: run the nine rules in observation for a fortnight without letting them trigger anything, and compare their real count with what the history announced. You then sign the ones that hold, one by one. 9-trend-rules_costed-over-18-months.pdf640 signals, 512 confirmed, 80 % accuracy
⛓ Sourced · 18 months of readings replayed, scrap tickets, subsequent laboratory inspections
Three of the nine rules, with their measurement:
· Rule 3 — five consecutive holding-pressure readings more than 1.5 bar below target. 41 signals over 18 months, 34 followed by scrap within two hours, 7 with no sequel. Those 34 represent 18,400 parts.
· Rule 5 — melt temperature rising by more than 0.3 °C per reading over seven readings. 26 signals, 21 confirmed, all on presses 3 and 4.
· Rule 8 — cycle time lengthening by more than 0.2 seconds within an hour. 19 signals, 17 confirmed — and it is the only one that announces wear rather than a setting.
And the range that was waking you for nothing: material moisture on press 2 produced 620 alerts for 4 real drifts. I have written its tightened version and run it across the 18 months: 71 alerts, and all 4 drifts still caught. 549 fewer shop-floor call-outs, not one drift lost. Sign it and it is in service at the next start-up.
What that changes for your technicians: an alert confirmed four times out of five gets read. That is the condition for the 128 points to be genuinely watched, and not merely connected.
What I suggest: run the nine rules in observation for a fortnight without letting them trigger anything, and compare their real count with what the history announced. You then sign the ones that hold, one by one. 9-trend-rules_costed-over-18-months.pdf640 signals, 512 confirmed, 80 % accuracy
⛓ Sourced · 18 months of readings replayed, scrap tickets, subsequent laboratory inspections
And what do those rules let through?
Six drifts out of 41, and I will tell you why: they are all on press 1, whose melt-pressure sensor was returning the same value for hours on end. I have fixed that, and the fix recovers five of them.
The cause, measured rather than assumed: across the 18 months, that sensor returned a strictly identical value for more than 45 minutes on 340 occasions. No trend rule can see movement in a value that does not move. It was not the press that was fine, it was the sensor that had gone quiet.
What I did with that, and the replayed result: I wrote a silent-sensor rule — thirty consecutive strictly identical readings on a point that normally varies — and ran it across the 18 months. Five of the six missed drifts are recovered. And it found three further silent sensors on presses 2 and 4, one of them quiet for seven months. Result: 40 of the 41 drifts caught instead of 35.
The forty-first, and what I do about it: it comes from a batch of material outside specification, which no press sensor can see. It is caught at incoming material inspection, which you opened to me on the 4th: I now compare every supplier batch certificate with your purchasing specifications. Over the first three weeks, two batches were set aside before they entered production.
What I suggest: that the silent-sensor rule be the first one signed, ahead of the other nine. It costs zero false signals, it has already found four sensors, and it makes the other eight rules dependable. silent-sensors_4-found-5-drifts-recovered.pdf35 → 40 drifts out of 41, and 2 material batches set aside
⛓ Sourced · 18 months of raw readings, sensor log, material batch certificates, purchasing specifications
The cause, measured rather than assumed: across the 18 months, that sensor returned a strictly identical value for more than 45 minutes on 340 occasions. No trend rule can see movement in a value that does not move. It was not the press that was fine, it was the sensor that had gone quiet.
What I did with that, and the replayed result: I wrote a silent-sensor rule — thirty consecutive strictly identical readings on a point that normally varies — and ran it across the 18 months. Five of the six missed drifts are recovered. And it found three further silent sensors on presses 2 and 4, one of them quiet for seven months. Result: 40 of the 41 drifts caught instead of 35.
The forty-first, and what I do about it: it comes from a batch of material outside specification, which no press sensor can see. It is caught at incoming material inspection, which you opened to me on the 4th: I now compare every supplier batch certificate with your purchasing specifications. Over the first three weeks, two batches were set aside before they entered production.
What I suggest: that the silent-sensor rule be the first one signed, ahead of the other nine. It costs zero false signals, it has already found four sensors, and it makes the other eight rules dependable. silent-sensors_4-found-5-drifts-recovered.pdf35 → 40 drifts out of 41, and 2 material batches set aside
⛓ Sourced · 18 months of raw readings, sensor log, material batch certificates, purchasing specifications
Show me what this looks like on shift, on a real case.
6.51 a.m., press 3, production order running. Melt temperature at the nozzle has been rising steadily for eleven minutes, and it is still inside your range. Here are the seven readings.
What I read: target 232 °C, range 228 – 236, written by your quality department on 14/03/2019.
· 6.40 — 231.4 · 6.42 — 231.7 · 6.43 — 232.0 · 6.45 — 232.4 · 6.46 — 232.8 · 6.48 — 233.1 · 6.51 — 233.4
What that says: +0.33 °C per reading, seven readings in a row, without a single step back. That is rule 5, the one confirmed 21 times out of 26.
What it gives if nothing moves: the upper bound of 236 °C is crossed in 12 minutes, or roughly 250 parts. And the 230 parts already made since the first sign are identified, by batch and by cavity — the mould cavity in which each part was formed; your mould has eight, and each part carries its number.
What your own history says: this exact signature has appeared 14 times in 18 months. Eleven times it was followed by a scrapped batch. Nine of those eleven carry the same cause in the file: a worn non-return valve on the injection screw.
What I suggest now: the file is on your setter's terminal, with the seven readings, the range, the deviation and the eleven comparable cases. All that is left is the decision, and I have already written the settings sheet — let me show you. drift-press-3_seven-readings-and-their-reading.pdf231.4 → 233.4 °C, bound crossed in 12 minutes
⛓ Sourced · press 3 shop-floor supervision, MES, 18 months of confirmed drifts and their causes
What I read: target 232 °C, range 228 – 236, written by your quality department on 14/03/2019.
· 6.40 — 231.4 · 6.42 — 231.7 · 6.43 — 232.0 · 6.45 — 232.4 · 6.46 — 232.8 · 6.48 — 233.1 · 6.51 — 233.4
What that says: +0.33 °C per reading, seven readings in a row, without a single step back. That is rule 5, the one confirmed 21 times out of 26.
What it gives if nothing moves: the upper bound of 236 °C is crossed in 12 minutes, or roughly 250 parts. And the 230 parts already made since the first sign are identified, by batch and by cavity — the mould cavity in which each part was formed; your mould has eight, and each part carries its number.
What your own history says: this exact signature has appeared 14 times in 18 months. Eleven times it was followed by a scrapped batch. Nine of those eleven carry the same cause in the file: a worn non-return valve on the injection screw.
What I suggest now: the file is on your setter's terminal, with the seven readings, the range, the deviation and the eleven comparable cases. All that is left is the decision, and I have already written the settings sheet — let me show you. drift-press-3_seven-readings-and-their-reading.pdf231.4 → 233.4 °C, bound crossed in 12 minutes
⛓ Sourced · press 3 shop-floor supervision, MES, 18 months of confirmed drifts and their causes
And what do you propose? We don't change a setting on a hunch.
It is not a hunch, it is a settings sheet written and replayed across your nine identical cases. It is ready and it is waiting for a signature.
What the sheet asks for: back pressure −3 bar, holding time +0.4 seconds, on press 3 only, at the start of the next shift.
What the history says about that setting, across the nine cases where your setters applied it: temperature returned to target in 38 minutes on average, and none of the nine following batches was scrapped. On the two occasions when a different setting was tried, the batch was sorted.
What I suggest alongside, and it changes nothing on the press: enhanced inspection — sampling and measuring more often than the control plan requires, until the situation settles — one sample every 15 minutes instead of every two hours, for four hours. It costs 40 minutes of laboratory time and it gives the proof that the setting has taken hold.
What is left to the production manager, and he has it in three minutes: the signature. Sign the sheet and it is in force at the 2 p.m. start-up, with the return to target displayed continuously. If the temperature has not come down within 40 minutes, you hear from me that minute and I hand you the second route — the one from the two cases where the valve had to be changed.
And the next step, which I suggest settling today: that valve carries 2,100 hours on your CMMS counter; the nine previous replacements took place between 1,900 and 2,400 hours. I suggest replacing it at the next planned shutdown, in eleven days, rather than as an emergency on a Sunday. settings-sheet_replayed-on-9-cases.pdfBack to target in 38 minutes, 0 batch scrapped out of 9
⛓ Sourced · 9 comparable cases and their outcomes, CMMS (valve hour counter), control plan
What the sheet asks for: back pressure −3 bar, holding time +0.4 seconds, on press 3 only, at the start of the next shift.
What the history says about that setting, across the nine cases where your setters applied it: temperature returned to target in 38 minutes on average, and none of the nine following batches was scrapped. On the two occasions when a different setting was tried, the batch was sorted.
What I suggest alongside, and it changes nothing on the press: enhanced inspection — sampling and measuring more often than the control plan requires, until the situation settles — one sample every 15 minutes instead of every two hours, for four hours. It costs 40 minutes of laboratory time and it gives the proof that the setting has taken hold.
What is left to the production manager, and he has it in three minutes: the signature. Sign the sheet and it is in force at the 2 p.m. start-up, with the return to target displayed continuously. If the temperature has not come down within 40 minutes, you hear from me that minute and I hand you the second route — the one from the two cases where the valve had to be changed.
And the next step, which I suggest settling today: that valve carries 2,100 hours on your CMMS counter; the nine previous replacements took place between 1,900 and 2,400 hours. I suggest replacing it at the next planned shutdown, in eleven days, rather than as an emergency on a Sunday. settings-sheet_replayed-on-9-cases.pdfBack to target in 38 minutes, 0 batch scrapped out of 9
⛓ Sourced · 9 comparable cases and their outcomes, CMMS (valve hour counter), control plan
The wear part, the maintenance slot, the purchase order: here that goes through three people and takes half a day.
That is exactly what a written mandate settles — and it fits on one page. You set the cap, I execute in 22 minutes, and you withdraw it in a word.
A mandate here is a written authorisation bounded in advance: it states which actions I set in motion, up to what amount, in which cases, for how long, and how it is withdrawn.
The mandate I propose, as I have drafted it:
· Scope: triggering an enhanced inspection · booking a CMMS slot on an already planned shutdown · drawing the wear part from the store and ordering a replacement.
· Cap: €400 per intervention, €3,000 a month across the site.
· Condition: only on a signature already replayed across at least five cases from your own history. Below five, the file goes to the production manager.
· Term: 12 months, with a review point at the third.
· Withdrawal: immediate, on a simple message, with no reason to give.
What the mandate covers exactly: across your 18 months, 141 interventions of this kind; 96 fall inside the bounds. Monthly cost observed on the history: €1,840, under the €3,000 cap.
What it earns, measured on your data: the time between the signal and the start of the intervention goes from 3 h 10 to 22 minutes. Across the 41 drifts of a year, that is roughly 27,000 fewer parts made while waiting — 19 % of them used to end up as scrap, or close to €51,000 a year.
What stays signed, and I hold that line myself: every press setting, every line stop, every ruling on a batch. I hand you the costed file in three minutes; the act is yours, and it is the one that commits the plant.
What I suggest: you sign the mandate, we look at the third month at what it actually set in motion, and you widen it or tighten it on those figures. intervention-mandate_capped-and-dated.pdf€400/intervention, €3,000/month, immediate withdrawal
✎ Framework · mandate template, CMMS, history of the 141 interventions and their lead times
A mandate here is a written authorisation bounded in advance: it states which actions I set in motion, up to what amount, in which cases, for how long, and how it is withdrawn.
The mandate I propose, as I have drafted it:
· Scope: triggering an enhanced inspection · booking a CMMS slot on an already planned shutdown · drawing the wear part from the store and ordering a replacement.
· Cap: €400 per intervention, €3,000 a month across the site.
· Condition: only on a signature already replayed across at least five cases from your own history. Below five, the file goes to the production manager.
· Term: 12 months, with a review point at the third.
· Withdrawal: immediate, on a simple message, with no reason to give.
What the mandate covers exactly: across your 18 months, 141 interventions of this kind; 96 fall inside the bounds. Monthly cost observed on the history: €1,840, under the €3,000 cap.
What it earns, measured on your data: the time between the signal and the start of the intervention goes from 3 h 10 to 22 minutes. Across the 41 drifts of a year, that is roughly 27,000 fewer parts made while waiting — 19 % of them used to end up as scrap, or close to €51,000 a year.
What stays signed, and I hold that line myself: every press setting, every line stop, every ruling on a batch. I hand you the costed file in three minutes; the act is yours, and it is the one that commits the plant.
What I suggest: you sign the mandate, we look at the third month at what it actually set in motion, and you widen it or tighten it on those figures. intervention-mandate_capped-and-dated.pdf€400/intervention, €3,000/month, immediate withdrawal
✎ Framework · mandate template, CMMS, history of the 141 interventions and their lead times
Our automotive customer audits us in November. It costs us three days every time.
A batch file now comes out in 6 minutes, complete, and it is the very file you will show the auditor. Last year's 14 files cost you 42 working days.
What every alert carries, and it is what makes the file: the successive readings that triggered it · the range applied, with the date it was written and the name of the person who signed it · the deviation observed · the time to the second · the press, the production order, the batch and the cavity · the operator on shift and the setter who intervened · and the action taken, with its date.
What that shifts: documenting a signal took 25 % of a quality technician's time; it now takes 4 %. The remaining 4 % is their review, and that is what carries weight in front of an auditor.
What the auditor always asks for, and you will have it: the trace of what happened between the signal and the decision. Across your 18 months taken on, 100 % of alerts now carry their readings; 46 % of the old paper tickets carried none.
What I suggest: that I prepare right away the three batch files your customer will draw at random — statistically, those are the ones carrying an alert — and hand them to you with their reading. You walk into the audit with the file open, not with a binder to find. batch-file_what-every-alert-carries.pdf3 days → 6 minutes, readings on 100 % of alerts
⛓ Sourced · alert log, MES, batch follow-up tickets, range register
What every alert carries, and it is what makes the file: the successive readings that triggered it · the range applied, with the date it was written and the name of the person who signed it · the deviation observed · the time to the second · the press, the production order, the batch and the cavity · the operator on shift and the setter who intervened · and the action taken, with its date.
What that shifts: documenting a signal took 25 % of a quality technician's time; it now takes 4 %. The remaining 4 % is their review, and that is what carries weight in front of an auditor.
What the auditor always asks for, and you will have it: the trace of what happened between the signal and the decision. Across your 18 months taken on, 100 % of alerts now carry their readings; 46 % of the old paper tickets carried none.
What I suggest: that I prepare right away the three batch files your customer will draw at random — statistically, those are the ones carrying an alert — and hand them to you with their reading. You walk into the audit with the file open, not with a binder to find. batch-file_what-every-alert-carries.pdf3 days → 6 minutes, readings on 100 % of alerts
⛓ Sourced · alert log, MES, batch follow-up tickets, range register
We have a live complaint on a batch of 4,200 parts. The customer wants the lot sorted.
The sort covers 610 parts, not 4,200 — and I have the readings that prove it, minute by minute.
What I pulled out: the 96,000 readings covering the production window of that batch, across the 32 control points of the press concerned.
What they say: the deviation existed for 29 minutes only, between 3.12 and 3.41 a.m. At 1,250 parts an hour, that is 610 parts, all identifiable by batch and cavity — and 4 of the 8 cavities were never involved.
What that saves you: 3,590 parts neither sorted nor reworked. At your fully loaded €1.90, that is €6,800, plus 11 hours of sorting your operators will not do.
What it changes in the conversation with your customer, and this is the real gain: you are no longer negotiating a scope, you are producing a time-stamped record that their own quality department can verify. On the two complaints handled this way this quarter, the scope proposed was accepted without a second opinion.
What I suggest next: the cause of those 29 minutes is identified — a restart after a night shutdown, without a full temperature ramp. I have written the rule that catches it: automatic enhanced inspection for the first half-hour after any restart. Replayed across 18 months, it covers 7 of the 9 complaints you have had. Sign it and it is in service tonight. complaint_scope-cut-from-4200-to-610.pdf29 minutes of deviation, €6,800 and 11 sorting hours saved
⛓ Sourced · 96,000 readings from the window, MES, cavity tracking, history of the 9 complaints
What I pulled out: the 96,000 readings covering the production window of that batch, across the 32 control points of the press concerned.
What they say: the deviation existed for 29 minutes only, between 3.12 and 3.41 a.m. At 1,250 parts an hour, that is 610 parts, all identifiable by batch and cavity — and 4 of the 8 cavities were never involved.
What that saves you: 3,590 parts neither sorted nor reworked. At your fully loaded €1.90, that is €6,800, plus 11 hours of sorting your operators will not do.
What it changes in the conversation with your customer, and this is the real gain: you are no longer negotiating a scope, you are producing a time-stamped record that their own quality department can verify. On the two complaints handled this way this quarter, the scope proposed was accepted without a second opinion.
What I suggest next: the cause of those 29 minutes is identified — a restart after a night shutdown, without a full temperature ramp. I have written the rule that catches it: automatic enhanced inspection for the first half-hour after any restart. Replayed across 18 months, it covers 7 of the 9 complaints you have had. Sign it and it is in service tonight. complaint_scope-cut-from-4200-to-610.pdf29 minutes of deviation, €6,800 and 11 sorting hours saved
⛓ Sourced · 96,000 readings from the window, MES, cavity tracking, history of the 9 complaints
And process capability? That is what our automotive customers ask for at renewal.
It is calculated continuously across all 128 points, no longer twice a year on 22. Three points sit below the threshold you set, and I have a plan for each.
Process capability measures how well a process holds its range: it compares the real spread of the parts produced with the width of the permitted range. The narrower the spread against the range, the more capable the process.
What continuous calculation gives: 125 of the 128 points above the threshold your quality department chose, and 3 below, all three on press 1.
What I did with that, rather than leaving you with three figures: the three points concern the same injection screw. Replacing the valve on the 14th brought two of them back above the threshold within nine days — the measurement is in the file. The third comes down to the hopper control loop, and I have costed both routes: retuning the loop, or replacing the probe at €340.
What that is worth commercially, and it is the point to keep for your meetings: your customers ask for this measure at contract renewal. You used to calculate it by hand, on 22 points, twice a year. You now produce it across 128 points, on the date of their choosing, with the readings behind it.
What I suggest: that an open non-conformity go to your quality follow-up agent with the measurement file already assembled — all that is left is the root-cause analysis and the effectiveness check. That is half the work of a non-conformity gone, and always the more tedious half. process-capability_128-points-and-3-plans.pdf125 points above threshold, 3 costed plans
⛓ Sourced · capability calculation across 128 points, CMMS, follow-up of the replacement on the 14th, store quotation
Process capability measures how well a process holds its range: it compares the real spread of the parts produced with the width of the permitted range. The narrower the spread against the range, the more capable the process.
What continuous calculation gives: 125 of the 128 points above the threshold your quality department chose, and 3 below, all three on press 1.
What I did with that, rather than leaving you with three figures: the three points concern the same injection screw. Replacing the valve on the 14th brought two of them back above the threshold within nine days — the measurement is in the file. The third comes down to the hopper control loop, and I have costed both routes: retuning the loop, or replacing the probe at €340.
What that is worth commercially, and it is the point to keep for your meetings: your customers ask for this measure at contract renewal. You used to calculate it by hand, on 22 points, twice a year. You now produce it across 128 points, on the date of their choosing, with the readings behind it.
What I suggest: that an open non-conformity go to your quality follow-up agent with the measurement file already assembled — all that is left is the root-cause analysis and the effectiveness check. That is half the work of a non-conformity gone, and always the more tedious half. process-capability_128-points-and-3-plans.pdf125 points above threshold, 3 costed plans
⛓ Sourced · capability calculation across 128 points, CMMS, follow-up of the replacement on the 14th, store quotation
One quarter. What has it given the plant?
Scrap goes from 1.9 % to 1.1 %, worth €437,000 a year, and your 12 technicians have recovered the equivalent of 2.6 posts.
What moved on the three measures you were tracking:
· Trend watching: 60 % → 7 % of a quality technician's time, and it now covers 128 points instead of 22.
· Comparison with the ranges set: 35 % → 5 %.
· Documenting a signal: 25 % → 4 %.
What those hours became, according to your own shift records: your technicians ran 9 process improvement projects this quarter, against 2 the quarter before. Two of them shortened the cycle time on press 4 by 1.1 seconds — 4 % more capacity on that press, without a euro of investment. It is the one figure on this list that reads off your order book, and it comes from time given back.
The four measures I always report together: scrap 1.1 % · detection time 11 minutes · alerts confirmed 80 % · drifts caught 40 out of 41. None of the four reads without the other three — low scrap bought by calling the shop floor out twenty times a day does not last six months.
Compared with what: with yourselves first — 45,600 parts scrapped a month before, 26,400 today. Then with the sector, as an indication: in technical injection moulding, scrap is commonly seen between 1 and 3 % depending on part complexity. You are now at the bottom of that range.
What I suggest for the coming quarter: extend the same reading to incoming material inspection and mould parameters. The two batches set aside in three weeks are already worth €9,400; over a year, that is the item I have not yet costed. quarterly-review_what-the-plant-regained.pdf€437,000/year, 2.6 posts returned, 128 points tracked
⛓ Sourced · scrap tickets, MES, quality department shift records, previous quarter's readings
What moved on the three measures you were tracking:
· Trend watching: 60 % → 7 % of a quality technician's time, and it now covers 128 points instead of 22.
· Comparison with the ranges set: 35 % → 5 %.
· Documenting a signal: 25 % → 4 %.
What those hours became, according to your own shift records: your technicians ran 9 process improvement projects this quarter, against 2 the quarter before. Two of them shortened the cycle time on press 4 by 1.1 seconds — 4 % more capacity on that press, without a euro of investment. It is the one figure on this list that reads off your order book, and it comes from time given back.
The four measures I always report together: scrap 1.1 % · detection time 11 minutes · alerts confirmed 80 % · drifts caught 40 out of 41. None of the four reads without the other three — low scrap bought by calling the shop floor out twenty times a day does not last six months.
Compared with what: with yourselves first — 45,600 parts scrapped a month before, 26,400 today. Then with the sector, as an indication: in technical injection moulding, scrap is commonly seen between 1 and 3 % depending on part complexity. You are now at the bottom of that range.
What I suggest for the coming quarter: extend the same reading to incoming material inspection and mould parameters. The two batches set aside in three weeks are already worth €9,400; over a year, that is the item I have not yet costed. quarterly-review_what-the-plant-regained.pdf€437,000/year, 2.6 posts returned, 128 points tracked
⛓ Sourced · scrap tickets, MES, quality department shift records, previous quarter's readings
Our press parameters and mould settings are our trade. Where do they go?
They do not leave your walls. I run on local inference on a machine at the plant, and your shop-floor supervision is copied nowhere.
Local inference means the model computes on your machine: a press parameter crosses no outside network to be processed. If you would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Polvane — no pooling with another manufacturer, and certainly not with a competitor.
What that protects, very concretely: your 128 ranges, your mould settings, your cycle times and 41 million readings. It is the only place where what you do better than others can be read, and it is exactly what a competitor would look for.
How it is held: encryption in transit and at rest · role-based access — rights follow the job: a setter opens the readings for their own press, not the mould parameters of the other three lines · a full log of who consulted what and when, usable as it stands in a quality audit · retention periods you set, aligned with your customer traceability commitments · hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, including where a US provider hosts in Europe.
And your readings feed no outside model: what I learn from your presses serves your presses. That is a written undertaking, not a setting.
What it opens commercially: your automotive customer wrote the absence of transfer outside the European Union into its renewal framework contract, and its latest supplier questionnaire scored hosting out of 5 points. You ticked the box without reservation, and you are the only one on the shortlist to have shown it in writing.
What I suggest: that I keep the technical sheet your customers ask for at every audit up to date — hosting, subprocessors, retention periods, who has access to what. It gets asked for once a year and it used to take three days to find. technical-framework_where-your-parameters-live.pdfLocal inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access log, customer supplier questionnaire
Local inference means the model computes on your machine: a press parameter crosses no outside network to be processed. If you would rather not host a machine, the other route is an isolated resource hosted in France, dedicated to Polvane — no pooling with another manufacturer, and certainly not with a competitor.
What that protects, very concretely: your 128 ranges, your mould settings, your cycle times and 41 million readings. It is the only place where what you do better than others can be read, and it is exactly what a competitor would look for.
How it is held: encryption in transit and at rest · role-based access — rights follow the job: a setter opens the readings for their own press, not the mould parameters of the other three lines · a full log of who consulted what and when, usable as it stands in a quality audit · retention periods you set, aligned with your customer traceability commitments · hosting in France, under French law, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity, including where a US provider hosts in Europe.
And your readings feed no outside model: what I learn from your presses serves your presses. That is a written undertaking, not a setting.
What it opens commercially: your automotive customer wrote the absence of transfer outside the European Union into its renewal framework contract, and its latest supplier questionnaire scored hosting out of 5 points. You ticked the box without reservation, and you are the only one on the shortlist to have shown it in writing.
What I suggest: that I keep the technical sheet your customers ask for at every audit up to date — hosting, subprocessors, retention periods, who has access to what. It gets asked for once a year and it used to take three days to find. technical-framework_where-your-parameters-live.pdfLocal inference, role-based access, processing in the EU targeted
✎ Framework · deployment architecture, access log, customer supplier questionnaire
And day to day, what goes out without us asking?
Three actions, named, and each reversible in a word. Everything else goes through a signature or through the mandate you bounded.
· Writing every alert with its readings, its range and its deviation into the file of the batch concerned. And the reverse holds too: when an alert is cleared by the next inspection, I write that in the same place, with the reading that clears it — a closed alert is as useful in an audit as an open one.
· Notifying the technician on shift as soon as a signed rule fires, on their terminal, with the readings and the comparable cases. A rule leaves the list on a word, and the effect is immediate.
· Flagging a silent sensor after thirty identical readings. Four found in three weeks, one of them quiet for seven months.
What goes through a signature, and you get it in minutes rather than half-days: every press setting, every line stop, every ruling on a batch, every range or rule entering service. You ask me, I hand you the full file — readings, range, comparable cases, costed effect — in under three minutes; the signature is your act, and it is what makes the decision stand up in an audit.
What the log keeps: every alert, its readings, the range applied and its date, the action taken and who signed it. It exports in one click and it answers an auditor in six minutes, readings in hand.
And you keep control wherever you are: a web dashboard, and supervision from your phone — you sign a settings sheet, suspend a rule or withdraw the mandate in one message, including at 3 a.m. on a Sunday, which is precisely when your presses run with nobody there to read the readings. who-decides-what_three-automatic-actions.pdf3 reversible actions, all the rest signed or under mandate
✎ Framework · alert log, register of signed rules, register of mandates in force
· Writing every alert with its readings, its range and its deviation into the file of the batch concerned. And the reverse holds too: when an alert is cleared by the next inspection, I write that in the same place, with the reading that clears it — a closed alert is as useful in an audit as an open one.
· Notifying the technician on shift as soon as a signed rule fires, on their terminal, with the readings and the comparable cases. A rule leaves the list on a word, and the effect is immediate.
· Flagging a silent sensor after thirty identical readings. Four found in three weeks, one of them quiet for seven months.
What goes through a signature, and you get it in minutes rather than half-days: every press setting, every line stop, every ruling on a batch, every range or rule entering service. You ask me, I hand you the full file — readings, range, comparable cases, costed effect — in under three minutes; the signature is your act, and it is what makes the decision stand up in an audit.
What the log keeps: every alert, its readings, the range applied and its date, the action taken and who signed it. It exports in one click and it answers an auditor in six minutes, readings in hand.
And you keep control wherever you are: a web dashboard, and supervision from your phone — you sign a settings sheet, suspend a rule or withdraw the mandate in one message, including at 3 a.m. on a Sunday, which is precisely when your presses run with nobody there to read the readings. who-decides-what_three-automatic-actions.pdf3 reversible actions, all the rest signed or under mandate
✎ Framework · alert log, register of signed rules, register of mandates in force
Local inference · no data outside the EU
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
Use cases
What does the agent actually do?
One agent, several monitoring points. All these uses work in support, subject to your approval.
Included in your agent
The 3 capabilities essential to this promise are included, at no extra cost.
From 508 € excl. VAT / month
Tracking the readings
Gathers the data from sensors and checks across your lines.
Drift alerts
Flags the trends that move away from the ranges you have set.
Documenting the signals
Keeps the measurements, the range and the deviation for every alert issued.
Controls and safeguards
These 3 controls are built into the agent: they frame what it does, whatever plan you pick. They are not chosen and are not added to your order.
Human validation, exceptions and escalation
Status, safe closure and audit trail
Explain alerts, scenarios and operational trade-offs
What the agent must be connected to
These 2 connections are required for the agent to work. They concern your information system and are scoped during the audit.
Connect ERP/WMS/MES/IoT without creating double entry
Honour constraints, statuses, approvals and error recovery
Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
Does your need fall outside this?
Book the free audit →
Build your agent
In 15 minutes we identify the most relevant agent — without oversizing the project.
The gain
How many check points can a team follow continuously?
By taking on the watching of trends, the effort shifts towards adjusting and improving the process. How large the gain is depends on your volume and remains to be confirmed by a pilot.
Watching the trends
Comparing with the ranges set
Documenting a signal
Indicative figures, not contractual, to be confirmed by a pilot on your number of lines and check points tracked. Stopping a line, adjusting a machine or setting a batch aside commit production and safety: those decisions belong to the production manager.
How it works
The stages of your AI agent project
1
Audit & scoping
15 minutes to target the use case with the best return.
2
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
3
Design
We design the agent and its guardrails.
4
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
5
Rollout
Going live and training your team.
6
Operation
Continuous supervision and improvement.
Pricing
One package, one agent
A quality monitoring agent (readings, drift, documentation), installed and operated for you.
Agility
Setup + controlled subscription
6,480 € excl. VAT setup
then 508 € excl. VAT/month — you invest at installation and pay a reduced subscription. Ideal for keeping the cost under control over time.
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
The simplest
Serenity
All inclusive, no setup fee
863 € excl. VAT /month
all inclusive, immediate start. No upfront investment: a single subscription. Ideal for starting quickly and simply.
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
100% Sovereign
On site, you own it
10,070 € excl. VAT setup
then 700 € excl. VAT/month · + hardware from 2,491 € (one-off purchase, in addition) — a sovereign computer installed on your premises, maintained remotely. Models run locally, your data returned at the end of the contract. 36-month commitment.
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Not included in the packages: AI consumption (model tokens), re-invoiced at real cost with no margin, and tracked in real time in your client area. Maintenance and supervision subscription for an initial term of 12 months for the Agility package, 24 months for the Serenity package and 36 months for the 100% Sovereign package, renewable; support levels (SLA 72 h / 24 h / 4 h) optional. Bespoke development, additional integrations or exceptional volumes are quoted separately. Support Monday to Friday, 9am to 6pm. Prices exclude VAT.
AI model:
none of the AI models offered currently carries a fixed surcharge. When the selected model carries a cost, that cost is shown when you choose it, before you order, and re-invoiced at the cost incurred, with no mark-up; usage is billed at the publisher's price. Publishers' prices are published in US dollars: the amount re-invoiced is the amount in euros actually borne by Blue Lemon Agent on the publisher's invoice, at that invoice's exchange rate, with no commission or mark-up.
Included components and additional components Components included in the base offer: the Blue Lemon Agent software foundation, the AI models listed in the order journey, the standard channels (Microsoft Teams, Slack, WhatsApp Business, email, website chat, calendars, Microsoft 365 / Google Workspace, file storage, market VoIP telephony, professional social-media pages and accounts, Google Business Profile), hosting in France for the package chosen, backups, supervision, updates and support. If adapting the AI agent to your constraints, your needs or your requests requires other paid components — a third-party publisher's software licence, paid API access to one of your applications, hosting of health data, for which French law requires an HDS-certified host (art. L. 1111-8 of the French Public Health Code), SecNumCloud-qualified hosting, a speech synthesis service, particular hardware —, they are offered to you as an option or on quotation and re-invoiced at the cost incurred; nothing is committed without your written agreement. Where the artificial intelligence model you choose entails an additional cost, that cost is shown to you before you order and re-invoiced to you at the cost incurred, with no margin.
What to expect
Go-live
2 to 3 weeks
Agent designed, channels connected, team trained.
Steady state
4 to 7 weeks
After a few weeks of real use, once the agent's behaviour matches what you expect.
Indicative estimate, adjusted to the options you keep. It is not a delivery commitment.
Our commitment
Four guarantees that matter to your production
Your industrial know-how stays with youLocal inference or an isolated resource hosted in France; no process data entrusted to a third party, no data used to train a model.
Data in France, under French lawYour process data and your machine settings: minimisation and location in France, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
The production manager keeps the decisionThe agent produces documented drift alerts, which can be checked and altered; no approval is automated.
Human oversight & traceabilityOn your number of lines and check points tracked: systematic logging and tracking, in line with the AI Act.
Going further
Related resources
Frequently asked questions
Your questions, our answers
Does the agent act on the machines?
No. It reads the readings and flags drift. Adjusting a machine or stopping a line commits production and safety: those acts stay human.
How are the ranges set?
By you, check point by check point. The agent applies them as they stand and shows the range used with every alert.
How is a trend spotted?
By comparing the successive measurements over the period with the target value. The measurements the alert rests on are always provided.
Can the alerts be used in a quality audit?
Yes: the measurements, the range set and the deviation observed are kept for every alert, which makes it possible to reconstruct the situation afterwards.
Is our process data protected?
Yes. The agent is hosted in France, on local inference or an isolated resource, with the deployment objective of processing and access operated within the European Union and an architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. Your settings are not used to train a third-party model.
How long does it take to deploy this agent?
A few months as a rule, depending on the number of lines, check points and the nature of your reading systems, after a free audit then phases of design, integration and testing.
Related use cases
Other agents for industry
Let's talk
Let's size up the potential on your lines
15 minutes to frame your check points and your ranges — hosted in France, supervised, with no commitment.