We build Agents. You build the Future.
● Business offer — Industry & productionThe AI agent for industry: find the right procedure, keep production flowing
On a production line, lost time hides in the details: the procedure nobody can find, the drawing whose revision is no longer certain, the spare part impossible to identify, the customer calling back about a manual. Your AI agent absorbs those searches and repetitive answers. Hosted in France — on local inference or an isolated resource — your drawings, process sheets and know-how stay in the factory. The operator and the management keep the decision.
Updated on
Fictional simulation — not a client result
History: 3 similar non-conformities in 14 months — all linked to a batch of out-of-tolerance seals. Suggested action: check the batch number of the seal fitted.
⛓ Sourced · quality document management + non-conformity history
✎ Action · non-conformity record ready for review — quality approves
For an organisation established outside the European Union, the law of its own country applies; at its request, we put equivalent safeguards in place (human review before any action, a log of the criteria, information for the people concerned).
In a manufacturing company, a Blue Lemon Agent agent instantly finds the right procedure, the right drawing and the right revision, answers after-sales and customer questions, and automates repetitive tasks — technical documentation, support, supplier invoices, order administration. Data hosted in France; the AI model is called at its publisher, except with the Sovereign model option. With that option, your drawings, process sheets and know-how are never exposed to a foreign service, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity. The time saved goes back to production and quality. Live within a few weeks Your teams write to it from Microsoft Teams, Slack or their email, and your customers reach it on WhatsApp Business, your website chat or email — with no account to create and nothing to install. These connections are included in every plan, at no extra cost, within the number of connections set out in your agent's quote.
Reference points describing our offer, not results measured at a client. The scale of the gain is confirmed by a pilot on your own scope.
Indicative estimate, to be confirmed by a pilot on your real files (general terms of sale, article 4).
Three options, one agent
A document agent (procedures, drawings, quality, support), installed and operated for you. Choose according to how you work. Prices exclude VAT — annual subscription, the time it takes for the gains to settle in.
Setup + controlled subscription
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Hosting in France, a dedicated and isolated resource
All inclusive, no setup fee
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Hosting in France, managed end to end
On site, you own it
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
The uses of AI in industry & production
Here are the capabilities included in the agent, then the neighbouring agents that can be added to it. The agent carries out by itself what you authorise it to do; you decide what it does on its own and what it submits to you.
Technical documentation & quality
Instantly find the right procedure, the right drawing, the right revision, a quality record or a manual across the whole document management system.
After-sales & customer support
Answering recurring requests: part numbers, manuals, warranties, status of an order or a repair.
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.
Chasing unpaid invoices
Chasing late customer payments in multichannel sequences, without letting the company's cash flow slip.
Debt collection agent (multichannel sequences) from 307 € excl. VAT / month Debt recovery →Qualifying incoming requests
Qualifying quote requests and contact enquiries, and automatically scheduling sales or technical meetings.
Lead qualification / appointment booking agent from 312 € excl. VAT / month Qualification & appointments →HR & onboarding
Answering the teams' everyday questions (leave, job descriptions, safety) and supporting new operators as they settle in.
HR agent — employee support, onboarding and HR documents from 348 € excl. VAT / month Discover the agent →Tender responses
Finding the technical material, writing the submissions and assembling the documents to respond faster to tenders.
Tender response agent (drafting) from 354 € excl. VAT / month Tenders →AI agent for technical support & documentation
Technical support answers the same questions a hundred times, and the documentation always falls behind the code.
Technical support & documentation from 344 € excl. VAT / month Discover the agent →Quality tracking
A non-conformity is worth what is done with it: a clear finding, a cause looked for, a corrective action and a check that it worked.
Quality agent (non-conformity tracking) from 277 € excl. VAT / month Discover the agent →Level 2 customer support
A level 2 incident is rarely unresolved for want of expertise: it is resolved slowly because reconstructing the context takes time.
Tier-2 customer support agent (diagnosis) from 299 € excl. VAT / month Discover the agent →AI agent for invoice processing
Entering, matching, getting approval, paying: processing supplier invoices is a repetitive, high-volume process that invites mistakes.
Automated invoice processing agent (P2P) from 506 € excl. VAT / month Discover the agent →Purchasing and suppliers
Approaching several suppliers takes discipline: writing the same enquiry for each one, chasing those who are slow, then bringing offers in different formats onto a single basis.
Purchasing / supplier agent (quotes, follow-ups) from 259 € excl. VAT / month Discover the agent →AI sales & CRM agent
Every salesperson knows it: the CRM is never really up to date.
Sales agent (meeting notes, follow-ups, CRM) from 307 € excl. VAT / month Discover the agent →In 30 minutes we identify the most relevant agent — without oversizing the project.
See the agent at work
5 everyday 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.
Fictional simulation — not a client result
· The vernier caliper at station 4 has been out of calibration for 12 days. 340 parts have been checked with it since.
· The drawing in use on the shop floor for part TR-118 is at issue B. The current issue has been D since March.
· A material batch on hold pending its certificate has been consumed on two works orders, on 03 and 05/08.
· A supplier amended a specification on its product sheet on 21/07, with no variation to your framework order. morning-watch_4-flags.pdf4 flags · parts and orders concerned
⛓ Source · instrument register, drawings and issues, material status, supplier sheets
340 parts, but not 340 decisions: they fall into 4 works orders and 3 part numbers. The call is made by batch, not by part.
What I could establish without assuming anything:
· 112 parts are still in stock — they can be re-inspected.
· 146 have gone to two customers, with the despatch note numbers.
· 82 are built into sub-assemblies still on site.
One thing that matters, and reassures: the last calibration record, of 14/05, was conforming with a 2 µm drift against a 30 µm tolerance. That is not a guarantee, it is a serious indication — and it heads the file.
The status decision stays with the quality manager, and it gets signed. The file reaches them complete. metrology_340-parts-traced.pdf4 orders · 3 part numbers · location of each batch
⛓ Source · works orders, despatch notes, last calibration record
Routing follows your organisation: calibration and the held batch to the quality manager, because those are status decisions; the drawing issue to engineering and the shop supervisor, both together; the supplier specification to purchasing, copied to quality.
With a chase: 12 h on calibration and the held batch — every shift adds parts to the scope; 48 h on the rest. Then a weekly summary to the site director: by subject, never by operator.
And I do not stop at the flag: the file is built before it arrives. The out-of-interval instrument goes out with the list of works orders inspected since its last valid date and the exact scope of the parts concerned; the held batch with its movement history, hour by hour; the drawing issue with both drawings side by side and the dimensions that differ, highlighted; the supplier specification with the version before and the version after. The quality manager no longer opens an investigation: they read a file and decide.
What stays a decision, and must: stopping a line, ruling on a batch status, amending a record. A batch status carries the name of whoever pronounced it — that is what makes it traceable in an audit and enforceable towards a customer. You rule, the record goes out with your sign-off, within the minute.
And I read what you have opened to me, station by station, every access logged and withdrawable on a word: nothing leaves the site.
✎ Proposal · watch and chases to be configured — you set the thresholds
Fictional simulation — not a client result
A "usual" value entered on an inspection sheet becomes a declared measurement. It is not inferred from an average — and if it is wrong, the whole sheet loses its value, not just that line.
What I looked for instead, and found: the missing dimension is the bore diameter. The measuring-bench report of 05/08 at 14:12 contains it, for all five parts of the sample, under the works order number. It simply was not carried across.
What I suggest: attach the bench report to the sheet, with the file reference and the timestamp. The line is filled by its source, which is worth more than a manual entry — including a correct one.
One signature remains to be applied, and it is not mine. inspection-sheet_TR-118-0442.pdfLine recovered · timestamped source
⛓ Source · bench report of 05/08, works order TR-118-0442
On this batch, 112 parts are still in stock: the sample can be redone without touching the despatch plan, provided it is done tonight rather than at 7 am tomorrow.
The expensive part is not the missing measurement: it is noticing it as the lorry is being loaded.
A proposal, if you approve it: every inspection sheet is matched against its works order's bench reports at closure, not at despatch. An empty line whose measurement exists is filled by its source; an empty line with no measurement is flagged to the inspector, with the time remaining before despatch.
No sheet is signed, no batch is released, no value is estimated. Over the last six months, that matching would have recovered 31 lines out of 34 — and flagged the other three two days earlier. missing-lines_6-months.pdf34 lines · 31 recoverable from their source
✎ Proposal · matching at closure — the signature stays with the inspector
Fictional simulation — not a client result
What was shipped: despatch note BL-2026-3318 of 12/06, 240 TR-118 parts, from works order 0417, itself built with seals from batch 7741-B/2603.
What else that seal batch fed: three other works orders, 620 parts, delivered to four customers between 02/06 and 18/07. Two of them have reported nothing.
What I do not conclude: that the seal is the cause. A sealing defect can come from the seal, the tightening torque, the surface finish or assembly at the customer's end. All four are open, and traceability does not choose between them.
What would move the diagnosis forward, and I can prepare: a comparison of the tightening parameters recorded on the four works orders. Two of them ran on a different driver. traceability_batch-7741-B-2603.pdf4 orders · 860 parts · 5 customers
⛓ Source · despatch notes, works orders, material batches, tightening records
All four orders ran within tolerance — specified torque 18 Nm ± 2. But their distributions do not match:
· Orders 0417 and 0421, driver no. 2: mean 16.4 Nm, 18% of points below 16.5.
· Orders 0429 and 0436, driver no. 1: mean 18.1 Nm, tight distribution.
The two orders complained about are the driver no. 2 ones. That is one more coincidence, not proof — but it is checkable in an hour on the bench.
What I have written to turn it into proof or rule it out: the test protocol — thirty parts, fifteen per driver, same material, same operator, same sequence — and the conclusion criterion set in advance: if driver no. 2 is the cause, the expected difference in mean is 1.7 Nm, and it is measurable in an hour. The bench is free at 14:00.
Why the customer does not yet have a cause: a cause announced and then withdrawn costs more than a delay in answering. What they get instead holds better: an exact scope, a dated test, and a criterion that will say yes or no.
What I have prepared for them, and it holds today: the acknowledgement, the full traceability of their batch, the exact scope of the parts concerned, and a commitment to come back within 5 working days. A customer given the exact scope waits better than a customer promised a cause. tightening-comparison_4-orders.pdf2 drivers · distributions compared
✎ Support · pattern documented — the diagnosis stays with engineering
Fictional simulation — not a client result
Six gaps, and the most interesting is not the largest:
· Verne Composants billed €3,240 above the order: two lines added with no variation.
· Kane Traitement billed for work with no goods receipt at all.
· Halbran Matières billed the right price — but on a specification amended on 21/07, with no variation to your framework order. The price has not moved; the product has.
· Three minor gaps, roundings below the €5 threshold.
The third is the only one that costs nothing this month and can cost a lot later: a material whose specification changes without the framework order following means a product qualification that no longer covers what is delivered. invoice-check_6-gaps.pdf128 invoices · 3 kinds of gap
⛓ Source · 128 invoices, orders, receipts, supplier product sheets
What changed, line by line: the product sheet of 21/07 amends the stated hardness, from 72 to 68 Shore A. The rest is identical.
What your qualification file says: the 2023 sealing test was run at 72 Shore A, and the report names it as an input parameter.
What I do not know: whether 68 changes anything. That is a question for engineering, not for supplier accounting — and it is asked once, not at every delivery.
What I have prepared: the question to the supplier (is this a change or a sheet correction?) and the extract of the qualification file, for engineering.
And a proposal that goes further: I can compare every supplier product sheet against the one in force at qualification, and flag any parameter divergence — even when the price does not move. That is precisely the case where nothing raises an alert today: the invoice is right, the receipt is conforming, and only the product has changed. specification_Halbran_before-after.pdf1 parameter amended · qualification extract
✎ Action · supplier question ready · the technical view stays with engineering
Fictional simulation — not a client result
The measurement: bore on part TR-118, tolerance 18.000 to 18.040 mm, value measured at station 4: 18.040 mm. Instrument within its calibration period, measurement uncertainty stated on the certificate: 0.006 mm.
The point that had not yet been configured: your inspection sheet says « within tolerance » and does not say whether the limit is included, nor what is done with the instrument's uncertainty. So I did not declare conformity on my own — and the part is not waiting: it is set aside under its works-order number, the 11 other dimensions on the sheet are measured and conforming, and the bench report is attached.
Three ways of doing it. Limit included, conforming: the part ships today, but half the instrument's uncertainty lies outside the tolerance. Limit excluded, non-conforming: no risk of shipping out of tolerance, in exchange for good parts scrapped — and on an upper bore limit they rework badly. Limit included, but conformity declared only if the value stays within tolerance once the instrument's uncertainty is deducted: here 18.040 − 0.006 = 18.034 mm, so conforming, and the calculation is written on the sheet.
I recommend the third. It is the only one that decides on what the instrument can actually measure, and your calibration certificates already carry the uncertainty: the inspection sheet only has to pick it up, it calls for no new measurement.
You can also give me another one — for instance: limit included and conformity declared, with a confirmation measurement on a second instrument whenever the value touches the limit.
Shall I remember it for the next sheets? 41 measurements fell exactly on a tolerance limit over twelve months, 29 of them on this TR-118 bore. If you approve the rule, I will carry it onto the next sheets without bringing them to you one by one, and I will give you the count every month. dimension-on-the-tolerance-limit_three-ways.pdf1 measurement · 3 ways of doing it · 41 comparable cases
⛓ Sourced · station 4 measurement, TR-118 inspection sheet, instrument calibration certificate, 41 measurements on the limit over twelve months
The rule, in one sentence: a value on the tolerance limit is conforming if it stays within it once the instrument's uncertainty is deducted; the calculation and the calibration certificate number appear on the inspection sheet; the conformity signature stays with the inspector.
What I saw re-reading the 29 measurements of this bore, and it is worth more than the rule: they are not spread on either side of the tolerance. The process mean is 18.032 mm for a tolerance centred on 18.020 — off-centre by 0.012 mm towards the upper limit, and across the year's 340 parts none went below 18.018.
What it is worth: re-centring the tool setting by 0.012 mm brings the process back to the middle of the tolerance. Last year's 29 limit judgement calls disappear, and with them 29 sheets to be settled by an inspector, at 8 minutes each — 3 hours 52 minutes a year, and above all 29 fewer decisions to take under the pressure of a shipment.
The decision is yours: re-centring a setting touches the process, it goes through your approval and a qualification run. I have prepared the chart of the 340 measurements and the capability calculation that goes with it; I touch no setting.
What lets you go back on today's rule: it is entered in the inspection sheet under its name, with its date, the inspector who approved it and the measurement that gave rise to it. You remove it with one line.
✎ Rule approved and proposal · inspection sheet, 340 bore measurements over twelve months, instrument calibration certificate — the setting stays with production
Your case is not here? That is exactly what a 30-minute conversation is for. Discuss your project →
The stages of your AI agent project
Audit & scoping
30 minutes to target the use case with the best return.
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
Design
We design the agent and its guardrails.
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
Rollout
Going live and training your team.
Operation
Continuous supervision and improvement.
Why AI matters to industry — and why it hesitates
The technical information exists: it is in the document management system, the ERP, the drawings, the quality records. But it is scattered, sometimes out of date, and the right document often arrives too late. Yet the data involved — drawings, process sheets, patents, know-how — is at the heart of the company's competitiveness.
! The issue
The factory is caught between ever-rising demands on productivity and quality, and technical knowledge split between the document management system, the ERP, the drawings, the heads of long-serving staff and spreadsheets. Yet most consumer AI solutions amount to entrusting drawings, process sheets, bills of materials and industrial know-how to a third party, often hosted outside Europe and subject to the Cloud Act.
✓ Our answer
AI is only of interest to a manufacturer if it is sovereign and confidential by design. Local inference or an isolated resource hosted in France, systematic human oversight, decisions reserved to the teams: the time saved on finding information and on support is never paid for in exposed intellectual property. The aim is not to replace the operator or the design office, but to get them the right information, at the right moment.
Protecting drawings and know-how: sovereignty & compliance
A manufacturing company handles what it holds most precious: its drawings, its processes, its bills of materials. Here is how the architecture of our agents protects them, line by line.
Local inference
With the 100% Sovereign package, the agent runs on a machine in the factory: no drawing leaves the network, nothing passes through a cloud.
Hosting in France
Otherwise, with the Agility and Serenity packages, a dedicated, isolated resource hosted in France, under French law: data hosted in France; the AI model is called at its publisher, except with the Sovereign model option.
Reduced extraterritorial exposure
Exposure of drawings and know-how to the Cloud Act and FISA 702 is reduced by design; location alone does not guarantee immunity.
One isolated resource per site
No pooling of industrial data: an environment strictly dedicated to your company.
Encryption & controlled access
Encryption in transit and at rest, role-based access (RBAC), strong authentication and logging.
AI Act: governed deployment
The agent carries out what you have authorised; no action approved 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.
- The encryption mechanisms in transit and at rest, their components and key management are those documented for the architecture chosen.
- 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.
How much time can a factory win back?
By making technical documentation findable in seconds and absorbing repetitive answers, an industrial site can aim for a noticeable reduction in time lost searching for information and on support — reinvested in production and quality.
Four guarantees that matter to a manufacturer
Your questions, our answers
Can AI really find the right version of a drawing or a procedure?
Are my drawings and know-how protected?
Do we have to change ERP or document management system?
Can the agent take calls and handle customer support?
Does AI help with mandatory electronic invoicing?
Do you have to be a large group to equip yourself?
How long does it take to deploy an agent?
Which tools can people use to talk to the agent?
10 AI agents Blue Lemon Agent deploys for this scope
Other industries that produce and deliver
Let us estimate the potential at your site
30 minutes to identify the use case with the best return — hosted in France, supervised, with no commitment.