How it works

From source to answer.

Every source comes in through the same door, is normalised and indexed. The question decides which retrieval method to use: an exact query to the ERP or a search by meaning across the documents.

The full picture. One map of the business.

Your systems and your documents come in through the same door and become a living model of the business. On top of it, your team, your AI and your agents ask questions, each with its own permissions.

KnowOS building blocksYour sourcesWho asksAgents with their own tokenLiving model of the businessDigital Sage
OntologyThe map of the business
Digital SageAnswers with a source
HypothesesValidated patterns
Manual and FAQBuilt from your criteria
MCP serverFrom your own tools
AlertsCritical stock
AnomaliesOrders and margins
ReplenishmentDraft purchase orders
ReportsScheduled
ERP and CRMOrders, prices, customers
Integrafy DataThe PIM: attributes and statuses
DocumentsPrice lists, manuals, email
PeopleBrowser, Teams, Slack
Your AIChatGPT, Claude, Gemini
AppsREST and GraphQL API

Step by step

Four steps, without changing what already works.

  1. 01 · Capture

    Connect the systems and documents you already use.

    ERP and PIM read-only, price lists in PDF and spreadsheets, processes and manuals, and email and calls turned into criteria.

  2. 02 · Structure

    A living model of the business: the ontology.

    Your ERP speaks in tables. Your business speaks in customers, orders and agreements. The ontology translates: named objects, their relationships, their actions and who can see what.

  3. 03 · No contradictions

    If two sources don't match, the system flags it.

    It doesn't choose for you. It marks the conflict, alerts the owner and records the decision. Every criterion keeps its origin and how long it applies.

  4. 04 · Query

    Ask from wherever you work.

    In the browser, in Teams or Slack, or from Claude, ChatGPT or Gemini via MCP. Each person and each agent sees only what their token allows.

    For developers

In two seconds

What happens while the answer arrives. The model never touches the database.

  1. 01

    Question

    In natural language, with its token.

  2. 02

    Your AI

    Chooses the business tool.

  3. 03

    Permissions

    Role, rows, columns and actions.

    Token validation
  4. 04

    Ontology

    SQL and semantic search, already scoped.

  5. 05

    Answer

    With its sources. Any action stays a draft.

The rule that governs everything

The corpus answers knowledge. The ERP answers data.

If no criterion is recorded, the system says there is no record. It never improvises. And from a twelve-minute call, only the criterion it contains is kept.

  • Nothing is made up: every answer cites its source and its date.
  • Inherited from the source: if you can't see the margin in the ERP, you can't see it by asking either.
  • Always correctable: any criterion can be adjusted, withdrawn or replaced.
  • Conflicts don't resolve themselves: a person is alerted.
See security

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 request to IT. You leave with the scope of sources and a fixed price.