root@fr:~$ workspace
An agentic AI layer of your own, over your data and tools.
root Workspace puts agents to work over your documents and business apps, in France, on models you control end to end. Your teams keep the judgment and the liability; the AI does the repetitive work and cites its sources. Deployed on your side, as part of your guided adoption.
The product
Four pillars, one promise: you keep control.
What separates an AI layer you own from an assistant you rent: isolation, orchestration, traceability and governance are built in, not optional.
Per-customer isolation
Your own instance, your own stores, your own application. Never a shared multi-tenant database. Your data never crosses another customer's, and no one trains on it.
Your confidential documents stay yours, partitioned by design.
Multi-model orchestration, jurisdiction under control
The right model for each task, aware of the law that applies to it. The conductor chooses, routes and switches so you never depend on a single provider.
If access is cut, the work goes on: no process rests on a single point of failure.
Citations on every answer
Every answer points to the exact passages it rests on, inside your sources. No black box: a professional verifies in one click before putting a signature on the line.
A hallucination never becomes your liability: you see where every claim comes from.
Governance, guardrails and audit
A governance, guardrail and audit-log layer over self-hosted open models. Mandatory human oversight, per-person and per-team access rights enforced down to each answer, full traceability.
You prove who asked what, of which model, under which law: compliant by design.
Scheduled work
Your agents work on schedule. Nothing is written without you.
A Workspace agent does not wait to be spoken to. It starts every night, every Monday at 7am, or the moment an event lands in your systems. It reads, analyzes, prepares. And everything it wants to write into your tools waits for your go-ahead: in the morning, you review and release in one click.
On schedule or on event
A regular rhythm or a trigger wired to your systems: the work starts without anyone having to think about it.
Writes wait for your go-ahead
The agent prepares the update, the reply, the entry, then places it in your approval queue. Nothing lands in your tools without being reviewed and released.
Every run leaves a trail
A scheduled run is a conversation like any other: reviewable, sourced, logged. You see what the agent did, and on which documents.
That is our definition of autonomy: the agent executes, you keep the decision.
The conductor
Sovereignty is decided agent by agent, not in one blanket setting for the whole account.
Every agent defaults to our sovereign models: open-weight, hosted in France, with no US entity in the data path. For an agent with no sovereignty constraint, bring your own keys (BYOK) and point it at a frontier model. The sensitivity of the task decides, agent by agent.
| Agent | Model | Jurisdiction | Status |
|---|---|---|---|
| Drafting legal deeds | Sovereign model (open-weight) | France | + Sovereign |
| Contract analysis | Mistral | France (EU) | + Sovereign |
| Document search | Sovereign model (open-weight) | France | + Sovereign |
| Internal summary, non-sensitive | Frontier model (BYOK) | United States | ! Exposed, by choice |
Per-agent model choice + BYOK
You decide, agent by agent, which model runs which task. Sovereign by default; the frontier model on your own key when the task carries no sensitive data. No lock-in: you change models without changing product.
A living governance view inside the Workspace, continuously updated: per agent, which model, where, under which law. Visibility the platforms hosted elsewhere do not give you.
Connectors and integration
Wired into your business tools. Owned, not rented.
root Workspace connects to your documents, your inboxes and your business apps so the agents work on your actual business, not a perimeter cut off from everything else. The difference with the embedded AI of US SaaS suites: here you own the AI layer, not a vendor who bills it as an add-on.
The embedded AI of US SaaS suites
Your most sensitive operational data processed by an AI you do not own, on a US-controlled platform, billed as a premium add-on that deepens lock-in.
root Workspace
The same integration logic, but the AI layer is yours. Sovereign models by default, data that never leaves the territory, and the right to walk away with your agents.
Connectors and custom integrations built for your process, not a generic plug-in. And an open API: your systems and your existing AI tools can query your agents, under the same access rights.
Self-hosting
On-premise and self-hosted, for the most sensitive environments.
When sovereignty is critical, root Workspace runs on your own infrastructure or on certified French partner infrastructure. Data never leaves the territory, and CLOUD Act immunity is guaranteed by architecture, not by a promise.
On your own infrastructure
Deployed on your infrastructure or on partner infrastructure certified ISO 27001 and HDS, aligned with NIS2.
Immunity by design
No US entity in the data path: not the host, not the vendor, not the model.
Production-ready
Reliability and oversight from day one: if something fails, the service stays available and fails over cleanly, with no abrupt outage.
Put your first agent into production.
We map your processes, pick the first agent with you, and it is in production within weeks. From then on, you see at any moment which model runs, where, under which law.