The price does not come from the list
Often it comes from the last price that customer paid. Neither average price nor cost: the history and what was agreed.
Module 03 · Integrafy Platform · The decision operating system
KnowOSAsk, instead of opening seven tabs.
The knowledge that decides a sale and does not fit in an ERP field: PDF price lists, price history, agreed terms and internal procedures. Queryable in natural language, with the source cited.
The challenge
The ERP stores the transaction. It does not store the reasoning.
In most industrial and distribution companies, the commercial decision depends on information that lives in no system: it is in PDFs, in emails and in the experience of whoever has been there for years.
Price lists live in PDF and Excel
Readable by a person, not by a system. And they are the real source of much of the commercial terms.
Reports are requested and waited for
“I need all these contacts from the last five years” ends up as a SQL query, a ticket to IT and a queue.
The reasoning leaves with people
When someone key retires or leaves, their judgement disappears and there is no record of how decisions were made.
Negotiating off-list is an advantage as long as the reasoning is available when needed. It stops being one when it depends on a specific person picking up the phone.
Integrafy KnowOS
Not a dashboard. The reasoning layer that is missing.
It connects the company systems (ERP, PIM, files, documentation) and turns them into a living model of the business. It lets you understand what is happening, why, and what to do about it.
It does not replace the ERP or the PIM: it makes them queryable. And it does not replace the people who know: it multiplies their reach, so they stop repeating the same thing every day.
In one sentenceStop searching. Ask.
Where KnowOS sits in Integrafy Platform
KnowOS works on the knowledge base: the operational data synced from the ERP and the PIM, plus the company documentation. Natural-language conversation and agents sit on top.
How it works
From the source to the answer.
Every source enters through the same door, is normalised and indexed. The query decides which retrieval mechanism to use.
01 · Capture
SystemsERP and PIM, read-only
FilesPDF price lists and spreadsheets
DocumentationProcesses, manuals and regulations
ConversationsEmail and calls turned into criteria
KnowOS
02 · Structure · a living model of the business- Normalisation and indexing
- Ontology: entities and relationships
- Criteria with origin and validity
- Permissions inherited from the source
03 · Query
Digital sageNatural-language questions, with the source cited
HypothesesPatterns proposed and validated against the data
Manual and FAQGenerated from the recorded criteria
MCP serverThe knowledge, from your own tools
The rule that governs everything
The corpus answers knowledge. The ERP answers data. If there is no recorded criterion, the system says so.
It never improvises. And a twelve-minute call is not knowledge: what matters is the criterion that appears inside it for twenty seconds. The system looks for that and discards the rest.
Nothing is invented
Every answer cites its source and the date it was said. Without enough backing, the system says so.
Inherited from the origin
If someone cannot see the margin in the ERP, they cannot see it by asking either. Permissions are inherited from the source system.
Always correctable
Every criterion can be adjusted, withdrawn or replaced. When two sources contradict each other, the system does not choose: it flags it.
What really matters
The magic is not in the AI model. It is in the map you give it.
Models change every few months. The map of your business (what exists, how it relates, who can see what and what can be done) is what turns any model into someone who knows your company.
What an ontology is · explained without jargon
Your ERP speaks in tables. Your business speaks in customers, orders and agreements.
How the system stores it
Table · T_CLI_001
| CLI_ID | NOM_ABR | ORG_V | COND_P | GRP |
|---|---|---|---|---|
| 0000100234 | SUMIN.NORTE SL | 1000 | Z030 | 02 |
| 0000100871 | FERRET.CAMPS | 1000 | Z060 | 05 |
| 0000101102 | INST.LEVANTE | 2000 | Z030 | 02 |
| 0000101377 | ELECTR.RIBA | 1000 | Z090 | 07 |
Table · T_VEN_POS
| DOC_V | POS | MAT_ID | CANT | PR_NET |
|---|---|---|---|---|
| 4500012 | 010 | A-4471-B | 24 | 38.15 |
| 4500012 | 020 | A-0930-X | 6 | 112.40 |
| 4500019 | 010 | A-4471-B | 40 | 36.90 |
Codes and columns only their designer understands. An AI model reading this guesses.
How the ontology understands it
Objects with business names, connected to each other. The same data, now with meaning: an AI model reading this reasons.
The grammar of your company
An ontology is the language in which your business explains itself.
Data, logic, action and security in the same place. That is why the AI does not just find the data: it understands what it means and what can be done with it.
01
Objects · the nouns
Customer, order, product, price list, document, criterion.
02
Properties · the adjectives
What describes each object: market, payment terms, risk.
03
Relationships · the syntax
An order is placed by a customer and contains products.
04
Actions · the verbs
Create quote, approve discount, replenish stock. With rules and validations.
05
Permissions · who may say what
They travel with every object, every property and every action.
Customer
e.g. Suministros Norte, S.L.
source: ERP + CRM + email
Properties · what it is
NameMarketPayment termsCredit riskMargin · restricted
Relationships · what it connects to
→ Orders→ Price lists→ Documents→ Criteria→ Assigned rep
Actions · what can be done
Create quoteApprove discountBlock for risk
Permissions · who sees and who acts
Sales: own portfolioFinance: margin and riskAgent: read-only
Illustrative example · a “Customer” object with everything the AI needs to know about it
Relational + vector
What happened and what was said, on the same map.
The ERP knows how much a customer bought. Email knows what they were promised. Until now they were two worlds. The ontology anchors them to the same object.
Relational · what happened
ERPOrders · prices · stock
CRMAccounts · opportunities
WMSLocations · shipments
Exact query · SQL Ontology · one map
CustomerOrderCriterionDocumentProduct
Vector · what was said and why
Price lists and contractsPDF · spreadsheets
Manuals and datasheetsTechnical documentation
EmailAgreements and exceptions
Call audioTranscribed and chunked
Search by meaning · embeddings Every row and every document fragment is anchored to a business object. That is why one question can cross what happened with what was agreed.
One question
“What terms have we agreed with Suministros Norte and how much have they bought in the last 12 months?”
SQL · relational
Customer invoices in the ERP
€184,600 · +12% year on yearsource: ERP · salesSemantic · vector
Account email and contract
8% discount on family Asource: account manager emailOne answer
Both figures together, with both sources cited.
Endpoint · MCP · per-token permissions
Connect your AI, whichever it is. The ontology sets the rules.
The model never sees the database. It sees tools with business names, and only the ones its token allows it to use.
Who asks · people, models and agents
ChatGPTClaudeGeminiCopilot · TeamsOwn agentsn8n · ZapierInternal apps
One endpoint and one MCP serverREST API · GraphQL · MCP — the same contract for everyone
Switch models without touching the integrationPermission engineEvery request carries a token · the token decides the scope
Rolesales, finance, agent
Rowsonly their portfolio or area
Columnsmargin and cost hidden
Actionsread, propose or execute
OntologyThe tools the model sees have business names
find_customer()orders_of(customer)current_price(product, customer)search_documents(text)propose_discount() · actionRelational dataERP · CRM · WMS · PIM
Vector dataDocuments · email · audio
Audit · who asked, with which token, which tool they used and which data they received
Segmentation by token · illustrative example
The same question, three different answers. And all three are correct.
Each person talks to the part of the company that belongs to them. You do not build one assistant per department: there is one, and the token decides what each person sees.
The same question“How is Suministros Norte doing and what margin do they leave us?”
Finance director
- All customers
- Margins and costs
- Credit risk
Revenue of €184,600 and an 18.4% margin. Down two points due to off-list discounts on family A.
Sales rep, North
- Own portfolio, including this customer
- Sales and terms
- Margin and cost
Revenue of €184,600, 12% more than last year. Margin is not available for your profile.
Sales rep, South
- Own portfolio
- Customers from other areas
This customer is not in your portfolio. I cannot give you their data.
Step by step
What happens in the two seconds an answer takes.
Person with their token
Their LLM GPT · Claude · Gemini
MCP + permissions validates the scope
Ontology translates into queries
Data relational + vector
1 Question in natural language
2 Picks the tool and calls it with the token
3 Validates the token · resolves role, rows, columns and actions
4 Query already limited to what is allowed
5 SQL + semantic search
6 Records and fragments with source
7 Filtered, cited result
8 Answer with sources · if there is an action, as a draft
Audit · records every step · the model never touches the database: it only sees what the token lets it see
Agents under the same rules
An agent is just another user. With its token, its limits and its log.
The ontology does not distinguish between a person and an agent: both ask permission to read and to act. That is why agents can be connected without opening the database.
Example · replenishment agent token
- Reads stock, consumption and supplier lead times.
- Proposes purchase orders, always as drafts.
- Does not see margins, selling prices or customer data.
- Does not execute without approval from a person with permission.
AgentSame permission engine
- 01 · DetectsReads stock and consumption with its token
- 02 · ProposesPurchase order as a draft
- 03 · ApprovesA person with permission
- 04 · ExecutesWriteback to the ERP via Connect and Flow
- 05 · LogsObserve records what, who and when
Comparison
It is not KnowOS or your AI. It is your AI with a map or without one.
ChatGPT, Claude or Gemini can be used alone or connected to KnowOS through MCP. The difference is where the rules live.
Permissions Your AI directly · even on an enterprise planInherits those of each connected app (shared drive, SharePoint, email), app by app. If the ERP does not hide the margin from a rep, neither does the assistant. Your AI + KnowOSIts own permission engine on the ontology: role, rows, columns and actions per token. The same rule applies to every source and any model.
Session Your AI directly · even on an enterprise planCold. Every conversation starts from scratch; context is provided by uploading files or setting up a project. Memory belongs to each user, not the company. Your AI + KnowOSWarm. The ontology, the current criteria and the sources are always loaded and are the same for the whole company.
Structured data Your AI directly · even on an enterprise planWhatever fits in the conversation. Years of sales history get summarised, chunked or exported by hand. Your AI + KnowOSExact query on the full history. The model receives the result, not the whole table.
Reliability Your AI directly · even on an enterprise planIf it cannot find the data, it may complete the answer with what seems reasonable. Your AI + KnowOSIf there is no recorded criterion, it answers that it is not on record. Every answer cites its source and date.
Control Your AI directly · even on an enterprise planKnowledge lives in one vendor’s projects, memories and connectors. Changing models means redoing it. Your AI + KnowOSThe ontology and the corpus are yours and exportable. Switch models without touching the integration.
Limits Your AI directly · even on an enterprise planContext window, plan usage quotas and files per project. Cost grows with every document that is reloaded. Your AI + KnowOSThe model only receives what it needs for each question. Fewer tokens per answer and predictable cost.
Actions and audit Your AI directly · even on an enterprise planActions inside each connector. The audit is in the vendor console, separated by tool. Your AI + KnowOSBusiness actions with validation, draft and approval. One single log: who asked, with which token and what they received.
Company context
Every chat starts from zero. KnowOS never does.
With direct AI, each person supplies the context in each conversation. With KnowOS, the company context is already loaded, the same for everyone, and respects what each person may see.
Direct LLM · cold session
SalesTheir conversation
- pricelist_v3.pdf
- erp_export.xlsx
- email.txt
FinanceTheir conversation
- pricelist_v2_OK.pdf
- sales_2025.xlsx
- summary.docx
TechnicalTheir conversation
- pricelist_final.pdf
- datasheet.pdf
- notes.txt
Each person uploads their own version of the files. Every conversation starts from scratch.
What does not fit in the context window gets summarised or left out.
LLM + KnowOS · warm session
Sales
Finance
Technical
Whichever model each one prefers · GPT · Claude · Gemini
KnowOS · always loaded
OntologyCurrent criteriaPer-token permissionsCited sources
ERP · CRM · PIM · documents · email · audioOne context for the whole company. The model only pulls what it needs.
KnowOS architecture
Five pieces, one cycle that repeats.
Ingestion, Data Lake, models, ontology and AI that discovers patterns and generates validated hypotheses. No code and no technical data team.
Ingestion
ERP, eCommerce, CRM, CSV or APIs, no manual pipelines
Data Lake
Versioning and lineage built in, partitioned by customer and by time
Data models
Business objects that generate REST and GraphQL APIs
Ontology
Objects, relationships and actions, readable by AI via MCP
Hypotheses
Correlations detected and validated against real data
The cycle, without code
Connect→Model→Discover→Act
When a pattern is confirmed, it becomes an alert or a rule. And the cycle starts again.
Summary
The state of the business, at a glance.
- Consolidated metrics from ERP, PIM and files, with a single definition traceable to its origin.
- No duplicates or contradictions between systems: a single source of truth for management.
- Configurable alerts and access by roles and user profiles.
- Pending actions with a direct link to what needs solving.
Digital sage
The AI answers, explains and proposes.
- Not a generic AI: it sits on your business ontology, so it answers with your vocabulary and your data.
- “What price did we sell this reference to this customer at last time?”
- “Why has the margin dropped in this family?” and it explains the cause, not just the figure.
- “What would happen if we raised prices 5% in this segment?” a simulation before deciding.
- Every answer cites the source criterion, its date and the conversation it comes from.
Hypotheses
The system proposes. The data decides.
- Pattern detection on real data: revenue concentration, lead-time effect, discount thresholds.
- Every hypothesis is validated against the database and keeps its status: pending, validated or under review.
- Root-cause analysis when performance drops, instead of a chart that only says it fell.
- Early anomaly detection in sales, margins or stock.
Data sources
Everything that connects, in one place.
- ERP: items, price lists, customers and history. Read-only, no automatic writes.
- Integrafy Data (PIM): technical attributes, assets and completeness states. Comes in already normalised.
- Files: PDF price lists and spreadsheets, with extraction of tables and terms.
- Documentation: process diagrams, manuals and internal regulations.
- MCP server: the knowledge, queryable from your own tools. Not locked into yet another interface.
Use cases
What each area asks.
The same questions that today circulate by email, by phone or as a ticket to IT.
| Area | Typical question | Where the answer comes from |
|---|---|---|
| Management | “How has the average price of this family evolved over the last three years by market?” | ERP history crossed with archived price lists. |
| Sales | “What price and terms did we give this customer for this reference last time?” | Customer, product and price ontology, plus the email and price lists for that account. |
| Technical office | “Which references in this family still have the datasheet pending?” | Integrafy Data, with the completeness traffic light. |
| Marketing | “Which products are complete in German and ready to publish in the catalogue?” | Per-language and per-channel states in the PIM. |
| Operations | “What is the procedure to onboard a new supplier?” | Process diagrams and internal documentation. |
| New joiners | “I just joined. How do I process a sample for a new customer?” | Criteria base, without taking up a colleague’s time. |
Deliverables
What the team receives.
Living internal manual
Organised by area, with an owner and a review date. Exportable to PDF or to your intranet.
Internal FAQ
Generated from the recorded criteria and ordered by real query frequency, not by creation order.
Digital sage
In the browser and, if you prefer, in Teams or Slack. Every answer cites the source criterion and its date.
MCP server
The knowledge stays queryable from your own tools, including Integrafy Data.
Rules that act
Findings become actions on the connected systems: alerts, tasks and notices. Always as a draft for review.
The corpus is yours
The criteria and the knowledge base are yours and exportable in an open format. No vendor lock-in.
Security and data
How the information is handled.
Data processing agreement
Signed before touching a single file. It sets purpose, scope, retention period and the list of sub-processors with their location.
Nothing stays outside
Models work in zero-retention mode: the provider does not keep the submitted content or use it for training.
Dedicated server in the EU
Contracted in the customer’s name, with no intermediation or surcharge. The information stays within your perimeter.
Separated from the corpus
Pseudonymised before indexing, with the mapping table encrypted and stored separately.
Inherited permissions
By role and by source. What cannot be seen in the source system cannot be seen by asking either.
Complete log
Who asked what, what was answered and with which source. Every criterion with its origin and its last review date.
Rollout
Four phases. Read-only until the criteria are validated.
Diagnosis
- Sources and pilot domain
- Legal framework and processing agreement
Modelling
- Platform setup and system connection
- Business ontology
Capture
- Criteria and corpus construction
- Read connectors via MCP
Launch
- Agent, manual and FAQ
- Governance, permissions and audit
- QA, tuning and training
What we need from you
- A reference contact who knows the area, for occasional questions. No fixed time commitment.
- Access to the approved sources. Not one source more without your sign-off.
- A technical contact for export tests and the item master query.
- Signature of the processing agreement before capture begins.
Out of scope
- Language-model consumption above the contracted plan.
- Digitising paper documentation or documents not available electronically.
- Writing process documentation: KnowOS indexes what the company produces, it does not write it.
- Automatic writes to source systems: the first phase is read-only.
Roadmap
The foundation for what comes next.
The architecture is designed to evolve towards agents that operate proactively on the business. It is not what you contract at the start: it is what this phase makes possible.
Assisted financial management
Real-time control, cash alerts and profitability projections by family and market.
Assisted sales management
Advanced sales analysis, cross-selling opportunity detection and predictive pipeline management.
Assisted operations
Process optimisation, intelligent stock management and logistics efficiency.
None of this can be built on scattered data. The ontology and the criteria corpus are the prerequisite, and they are exactly what KnowOS delivers.
Plans
Three plans and one add-on. No surprises.
Monthly fee with no lock-in plus a one-off setup that includes ontology modelling and criteria capture. Prices exclude VAT.
Start
To start asking the ERP and the company documents.
190 €/month
Onboarding · €2,490 · ERP and Drive or OneDrive
- 3 GB of storage
- AI included: €20/month of usage
- 1 ERP connection
- 1 CMS: PrestaShop, Shopify, Sylius or Medusa
- 1 document connection, Drive or OneDrive
- Dedicated EU server included
Growth
RecommendedFor teams that query the ERP, the store and documentation daily.
390 €/month
Onboarding · €3,900 · includes 20 h of customization
- 10 GB of storage
- AI included: €109/month of usage
- 1 ERP connection
- 1 CMS: PrestaShop, Shopify, Sylius or Medusa
- 2 document connections, Drive or OneDrive
- Dedicated EU server included
Scale
For companies that cross ERP, CRM, store, documents and calls.
790 €/month
Onboarding · from €5,950 · with CRM, eCommerce and speech to text
- 50 GB of storage
- AI included: €109/month + about 25 M API tokens
- 2 ERP or CRM connections
- 1 CMS: PrestaShop, Shopify, Sylius or Medusa
- Multiple document connections
- Speech to text for calls and meetings
- Dedicated EU server included
Agents
KnowOS add-onAgents that read, detect and propose with their own token.
per agent
By scope, after the diagnosis
- Critical stock alerts
- Anomalies in orders, returns and margins
- Restocking suggestions
- Periodic automatic reports
- Actions always as drafts, with approval
- Same permissions and audit as a person
Setup is quoted by number of sources and pilot domain. Language-model consumption is at cost or with a monthly allowance. On-premise option on request.
See the full price list, packs and add-ons →Frequently asked questions about Integrafy KnowOS
Is Integrafy KnowOS a separate product or part of Integrafy Platform?
It is module 03 of Integrafy Platform, the knowledge layer. It works on the operational data synced by Connect, Data and Flow, but it can be contracted independently: read access to the ERP and to the company documentation is enough.
How is it different from connecting ChatGPT or Claude directly to my files?
In where the rules live. With direct AI, each person uploads their files and the context is lost when the chat closes. KnowOS keeps the ontology, current criteria and permissions always loaded, the same for the whole company, and every answer cites its source and date. The model can be swapped without touching the integration.
What happens if there is no recorded criterion for what I ask?
The system answers that it is not on record. It never completes an answer with what seems reasonable. And when two sources contradict each other, it does not choose: it flags it so a person can resolve it.
Can a sales rep see the margin by asking the AI?
Only if they can see it in the source system. Permissions are inherited from the source and applied per token: role, rows, columns and actions. The same question returns different answers to finance, to a rep in their own area and to a rep from another area, and all three are correct.
Do I need a data engineer or a technical data team?
No. Ingestion, the Data Lake, modelling and the ontology are configured without code during setup. If you have a technical profile they will appreciate the REST and GraphQL APIs and the MCP server, but it is not a requirement.
Where is my data processed and what does the model provider see?
A dedicated server in the EU contracted in your name. Models work in zero retention: they do not keep the content or use it for training. Personal data is pseudonymised before indexing, with the mapping table encrypted and stored separately. All under a processing agreement signed before touching a file.
Does KnowOS write to my ERP?
Not in the first phase: it is read-only until the criteria are validated. After that, actions (create a quote, propose a purchase order) are generated as drafts and executed by Connect or Flow after approval by a person with permission. Agents follow exactly the same rules.
Session on your case
Show us a question that takes an afternoon today.
A 30-minute session on a real case: a historical price list, an internal procedure, a query that today ends up as a ticket to IT. We leave with a diagnosis of the source scope and a fixed price.
Book a session