Our method
AI that speaks in your name has to be built differently.
C-OS is the engine shared by all our assistants. The eight rules below are built into it once and for all: no Cosensible assistant can step outside them.
- 8
- rules
- 0
- invented figures
- 1
AI only says what it knows
Our tools work from the company's data: sales, inventory, invoices, procedures, history. The language model phrases the answer; it doesn't decide the content. When data is missing, the tool says so and hands over, rather than filling in the gaps.
- 2
No figure goes out unchecked
Prices, quantities, totals, lead times: every number in an answer is compared with what was given to the model. A number from anywhere else gets the answer rejected; the model is told its error and tries again. Amounts are displayed straight from your data, never copied from a sentence.
- 3
The language of customers and teams, as they write it
Casual French, abbreviations, typos, English, Arabic, Darija, several languages in a single sentence: our tools understand what people really write, on a phone as well as in an email, not what they would write in a form. A glossary maintained with our clients, a search that understands even misspelled words, and an answer in the other person's language.
- 4
Each client's data is kept separate
Each company's data is stored on its own, with no access for other clients. Every tool is built on this rule, and we show on request how it's applied. One company, one space; nothing shared.
- 5
Your data stays yours
The data of our clients and of their own customers is hosted in France, on a server dedicated to Cosensible. Exchanges are encrypted (HTTPS), backups are daily, encrypted, and kept in the European Union, and your data can be exported at any time. Before any call to the AI, names and numbers are masked. We tell you in writing where your data is hosted and how it's processed, and we help you comply with your country's data protection law.
- 6
Costs that don't spiral
The risk with an AI tool is that its running cost exceeds what it brings in. We track what each answer costs and set a limit for each client: your subscription stays stable.
- 7
Humans stay in control
Our tools prepare, sort, check, and pass things on. They don't take payments, don't turn anyone away, and send nothing binding without approval. Every suggestion and every unanswered question is visible to your teams, who can correct it in one click.
- 8
Built into your tools, not bolted on
AI that forces your teams to switch tools goes unused. We build it into the software, files, and channels you already use, and we train your teams to use it and keep it in check.
What C-OS checks: illustrative example, fictitious figures
The figure is in your data: the answer goes out.
No 12-day payment time in your accounting data: answer rejected.
Every one of our tools is built on these eight rules, and we show on request how each one is applied. Together with our AI charter, they form the framework we hold ourselves to. We draw on international responsible AI frameworks, such as the NIST AI Risk Management Framework and the ISO/IEC 42001 standard, which we use as references, without claiming a certification we don't hold.
See the method at work.
Assessment, design, deployment, or audit: every engagement applies these rules, in every department.