root@fr:~$ solutions/dsi
Your teams already use AI. Without you. Take back control without slowing them down.
An agentic AI layer that IT governs: dedicated instance in France, models chosen and documented, access through your directory, full audit log, contractual reversibility. The official alternative that puts out shadow AI without putting out your teams' drive.
Governance is not a promise. It is a log line.
Every line says: which agent, which model and version, which data, which user, which validation.
The stakes
Shadow AI is not a risk to come. It is your starting point.
Usage came before policy
According to IDC (2025), 86% of French companies send sensitive data to non-sovereign AI tools. Prohibition does not remove that usage: it moves it to personal phones, beyond all measurement.
AI SaaS cannot be governed
Models that change without notice, data under foreign jurisdiction, logs missing or unusable: US-hosted platforms leave nothing to audit. You cannot govern what you cannot trace.
Control debt keeps piling up
Every tool adopted without review leaves shared credentials, forgotten OAuth grants, uninventoried flows. The bill comes due at the first incident, or the first security questionnaire from an enterprise client.
What you govern
An AI layer where you hold every parameter.
root Workspace deploys as a building block of your IT estate: dedicated instance, documented architecture, auditable behaviour. Not one more SaaS in the processing register: a layer the company owns.
Models chosen, versions pinned
You know which model runs, in which version, on which process. No silent updates: a model change is validated like an application release.
Dedicated instance in France
Your data stays on your instance, isolated per client, hosted in France. Nothing passes through a platform subject to the CLOUD Act, nothing trains a public model.
Access through your directory
SSO, role-based rights, per-agent scope: who accesses what is decided by your rules. An agent reads only what its scope allows, and every denial is traced.
Full audit log
Agent, model, version, data touched, user, validation: every action leaves a queryable trace, exportable to your SIEM.
Contractual reversibility
Data, configurations, history: everything exports in standard formats, with committed assistance and timelines. The exit clause is read before signing, not on the way out.
Predictable costs
An instance sized for your usage, not a consumption meter multiplied by your headcount. The AI budget becomes a line you can plan again.
Your IT estate stays yours
The AI layer plugs into what you run.
Your directory for identities, your tools for the work: mail, calendar, ERP, CRM, document management. Agents connect through APIs and MCP, an open standard. Nothing to migrate, nothing to replace. And if a tool changes tomorrow, the AI layer follows: it belongs to you.
The answer to shadow AI
You do not put out shadow AI with an internal memo. You put it out with a better alternative.
Your teams use consumer tools because those tools genuinely help them. Prohibition moves the usage, policy frames it, only a more effective official alternative replaces it. The method has three steps: measure what exists, set a short policy that can say yes, offer better. The detection checklist covers the first step.
Augment, not replace
The agent prepares. Your users decide.
No irreversible action without human validation: an agent prepares, proposes and documents, it signs nothing. Every agent has an access scope, a full trace and a kill switch. You switch off an agent the way you close an access: immediately, without breaking anything else.
CIO questions
What CIOs and CISOs ask us.
Which models run, and who decides on updates?
Open models, chosen with you task by task, versions documented in the log. No silent changes: a model update is validated like a production release, with rollback available. And the sizing is honest: we also tell you what each model cannot do.
Where does the data run, under which jurisdiction?
On your dedicated instance, hosted in France, under European jurisdiction, operated by root or by a regional infrastructure partner. No processing passes through a platform subject to the CLOUD Act. The exact location, the operator and the subcontracting chain are documented in the Sovereignty Map.
SecNumCloud, ISO 27001: where do you stand?
The hosting foundation is chosen against your requirement: root.cloud, or a regional partner's certified infrastructure when your specifications call for a specific qualification. In every case, the certifications of the chosen foundation are verified and documented with the supporting evidence: we never claim a qualification in the host's place.
How does the layer integrate with our IT estate?
Identities through your SSO and directory, API and MCP connectors to your tools, log export to your SIEM. The layer deploys as a building block of the estate, with its architecture documentation. One workshop is enough to check your exact configuration.
What exactly does the audit log cover?
Every action: which agent, which model and version, which data, which user, which validation, at what time. Including access denials. The log is queryable and exports to your SIEM. It is the one we show in demos: there is no sales version and real version.
How do we leave?
With your data, your configurations and your history, in standard formats, with contractually committed timelines and assistance. Reversibility is read before signing: it is a good test for any AI vendor, ourselves included.
The demo runs on your security questionnaire. Not on slides.
In one workshop, we go through your questions: models, jurisdiction, access, log, reversibility. You leave with documented answers, and an exposure reading of what you already run, shadow AI included.