AI that answers from your own knowledge. And hands over when it doesn’t know.
- Built for GDPR
- Your data never trains the models
- Data stored in the EU
- Human handoff
Knowledge Bases, Support Bots and Business Decisions for businesses of any size: answers for your customers and team, and decisions grounded in your own history.
- Answers from your own knowledge
- Hands over when it doesn’t know
- Your data never trains the models
- Data stored in the EU
- Odds, not certainty
A wrong answer costs more than no answer.
(01)What it does
Four ways in.
(02)Who it's for
Built for every business. That runs on its own knowledge.
- Your team answers the same questions every dayKnowledge Base
- Your customers wait for replies at 2 a.m.Support Bot
- You decide on gut feel while years of results sit in your systemsBusiness Decisions
Start with one pilot: we set it up with you, on your own data.
Explore features(03)Business DecisionsNew
Decisions grounded in your own history. with the odds stated.
Fast, structured decisions and return probabilities, for your business and inside it, drawn from your own history rather than the internet's.
- (01)
Calibrated probabilities
Each probability is checked against what actually happened to your past decisions, so 0.7 means seven in ten, on your record.
- (02)
Cited precedents
Every figure names the past decisions it rests on and how many there were. Nothing is asserted without its sources.
- (03)
Abstains when history is thin
Where too few precedents exist to state odds honestly, it says so and routes the decision to a person instead of guessing.
Decision brief
Open a second distribution hub in Lyon?
Horizon 18 months · P(return > 0)
Each coin is a past decision like this one: solid coins paid back, faint ones did not. The carriage stands at the odds, the collars clamp the range, and where the stack is too short the flag goes up.
For the business
The few large calls a year, where one mistake is expensive.
- Capital investments
- Market and site expansion
- Pricing changes
- Partner and supplier selection
Inside the business
The many routine calls a day, where consistency is the gain.
- Approvals and exceptions
- Credit limits for business customers
- Inventory and reorder levels
(04)How it works
Three steps. From your files to an answer.
Connect
Your documents, tools and past records, where they already live.
Ground
We clean and index them, so every answer comes from your own material. You choose what is connected.
Answer & decide
Answers for your team, replies for your customers, and Business Decisions with the odds stated. A person takes over when it isn’t sure.
(05)What we build
Core products. And how we set them up.
Your documents, cleaned and indexed in one place, so answers come only from what your company already knows.
- Planned
Fast, structured decisions and return probabilities drawn from your company's own past decisions and their outcomes. Calibrated, cited, and handed to a person when the history is thin.
Explore Business Decisions - Planned
An assistant for your customers that remembers the conversation, sends each question to the right place, and hands over to a person when it should.
We use enterprise model APIs, with zero retention where the provider offers it. Your data never trains them.
A one-off setup covers connecting your sources, cleaning the data and tuning the assistant to how your business works.
(06)Integrations
Seamless integration.
We connect the tools your documents, tickets and records already live in, as part of the setup, so nothing has to move.
(07)Models & data
Model-agnostic. By design.
- (01)
Leading model APIs, through one layer
Today Anthropic’s Claude writes the answers and Voyage AI turns your documents into a search index we keep in our own database, both through a single integration layer of ours.
- (02)
The right model for each task
Retrieval, support replies and decision briefs need different strengths, so each gets the model that suits it.
- (03)
Switch without a rebuild
When a better model arrives, it is swapped in behind the same layer. Your system, data and guardrails stay as they are.
- (04)
Your data never trains the models
It is sent only under enterprise API terms, with zero retention where the provider offers it.
PlannedComing next: MCP
Support for the Model Context Protocol, so the AI assistants and tools your team already uses can draw on your Crugon knowledge base, with the same permissions and guardrails.
(08)Method
How it’s built. And how it stays safe.
The Layers
Separate layers, each checked on its own: the panel you use, a gateway, retrieval from your sources, and the models underneath.
The panel is where your team and customers ask, and where answers arrive.
Before any model answers, retrieval finds the passages in your own material that the answer must rest on.
- Panel
- Gateway
- Retrieval
Where It Runs
Crugon is based in Turin, in the EU, and built around European data protection rules (GDPR).
The text of your documents and its index are stored in the EU, in Frankfurt. The AI models that read them run with their providers, which may be outside the EU.
You choose which sources are connected, and can disconnect them at any time.
- Turin HQ
- Built for GDPR
- EU data storage
When It’s Unsure
A wrong answer costs more than no answer.
Strict guardrails keep every reply to what your own sources say.
When it isn’t sure, it hands over to a person rather than guess.
- Human Handoff
- Strict Guardrails