Book a call

STEER · ASCENZIA SENTINEL

The control tower for your AI agents.

Every call your agents make goes through Ascenzia Sentinel, which lets through only what your rules allow and records each one in the execution log.

Hosted in France or Switzerland.

Demo data

STEER · USAGE

LEADERSHIP · CIO

See the AI nobody has declared.

Your teams are already using public AI tools, usually with the best of intentions and without mentioning it to anyone. Ascenzia Sentinel shows you which ones, and who is sending them what data; anything that should be blocked is.

AI usage flow · 30 days

All the tools run through Ascenzia Sentinel, which tells routine use apart from sensitive use and sets blocked attempts to one side.

AI usage flow · 30 days
TOOLSCONTROL TOWERTYPE OF DATAOpenAI Codex4,870 · 28%ChatGPT3,075 · 18%Mistral Le Chat2,950 · 17%Claude Code2,265 · 13%Gemini (Chrome)1,595 · 9.2%Claude965 · 5.5%Microsoft Copilot790 · 4.5%Perplexity575 · 3.3%Other tools330 · 1.9%Ordinary16,755 requestsSensitive share660 · 3.8%Blocked13 attemptsAscenzia Sentinel17,415 requests · 30 days
  • OpenAI Codex 4,870 requests (28%)
  • ChatGPT 3,075 (18%)
  • Mistral Le Chat 2,950 (17%)
  • Claude Code 2,265 (13%)
  • Gemini (Chrome) 1,595 (9.2%)
  • Claude 965 (5.5%)
  • Microsoft Copilot 790 (4.5%)
  • Perplexity 575 (3.3%)
  • Other tools 330 (1.9%)
  • Total 17,415 requests over 30 days: ordinary 16,755, sensitive share 660 (3.8%), 13 blocked attempts

Demo data

STEER · RULES

CIO · CISO

Nothing gets through without your rules.

You set the models and the guardrails, along with the list of tools each agent is allowed to touch. The agent works within those limits and stays there.

[01]

APPROVED MODELS

A catalogue of vetted models, from which each agent picks according to the type of task (a light model for sorting, a stronger one for drafting).

[02]

PER-REQUEST GUARDRAILS

They work both ways. Sensitive data is caught before it leaves, and a request that breaks the rules goes no further.

[03]

CONNECTORS AND ISOLATED EXECUTION

An agent opens only the tools listed in the registry, and it runs on its own, in an isolated environment.

STEER · COSTS

CFO · CIO

A budget in euros, capped in advance.

Each agent gets a maximum budget set in euros (your CFO doesn't think in tokens, and quite right too), fixed before its first request. An alert goes off before the cap is reached.

One gauge per agent, in euros, showing spend against the cap and flagging the alert threshold.

Budget per agentOctober1 alert
Spent
€927
Cap
€1,430
Alert threshold
80%
AgentSpentCap (€)Status
Municipal council agent€412 / €600
ok
Quote follow-up€188 / €250
ok
Document sorting€231 / €280
alert
Customer 360° view€96 / €300
ok

Demo data

PROOF · FRUGALITY

÷10

On simple tasks, the cost per token came down by a factor of 10.

-60%

fewer tokens after optimisation, on document-sorting agents.

We see no reason to pay for a top-tier model to sort invoices, so the gateway hands each task to the model that will do the job.

A fixed-price subscription, from €390 excl. VAT per month. AI consumption billed at cost, no margin, or covered by your own API keys.

See pricing

CHOOSE

LEADERSHIP · CIO · DPO

You change models when you decide to.

You take your models from a catalogue you set yourselves, and you decide which one each agent uses. When another model does the job better (or costs less), you switch without rewriting your agents or taking out a subscription with every vendor, while your contextual memory and your data stay on our platform, in France.

For each agent, the model you chose and its fallback. The fallback also covers you on the day a model goes down.

Multi-model gatewayAutomatic failover
AgentModelFallbackStatus
HR questionsClaude SonnetMistral Largeactive
Bank reconciliationMistral LargeClaude Sonnetactive
Meeting summariesGemini FlashMistral Smallfailover
TranslationMistral SmallGemini Flashactive

Demo data

EXECUTION LOG

Your agents' runs go into the log, with the model used, how long each took and what it cost in euros. You can export it on the day someone asks who did what, and with which model.

METADATA ONLY

Ascenzia Sentinel keeps metadata only, such as statuses, durations or tokens used. The text of prompts and responses is never collected.

REGULATORY FRAMEWORK

The platform helps you comply with DORA, NIS2, the AI Act, the Cloud Act and GDPR.

In France, in the Paris region, for our French clients. In Switzerland, with our partner Hidora, for our Swiss clients.

AVAILABLE TODAY

What is live today.

Nine features, available now, which we can show you on your own case from the first call. Everything listed here is already running.

  1. 01Multi-model gateway: routing, model changes whenever you decide, failover, consumption cap.

  2. 02Policies: catalogue of approved models, budgets per module, per source and per organisation.

  3. 03Traces and costs, tracked in euros.

  4. 04A capped budget for each agent.

  5. 05Agents created by your own teams, in Ascenzia Studio.

  6. 06Per-request guardrails, on input and output.

  7. 07Connectors held in a registry, and isolated execution.

  8. 08Shadow AI detection.

  9. 09An agent that prepares municipal council meetings, in production at a town hall in Seine-Saint-Denis.

YOUR QUESTIONS

‘We already have Microsoft Copilot.’

Keep Copilot. Ascenzia Sentinel sits alongside it with capped budgets and a catalogue of approved models, and lets you switch models without rewriting your agents.

‘What about the AI Act?’

We would not wait for the first inspection to get started. The job is to know which agents are running and what they cost, then to be able to show it; Ascenzia Sentinel keeps the execution log, costs in euros included, and you can export it whenever someone asks.

FAQ

Frequently asked questions

It is a platform where you create and steer AI agents while choosing where they are hosted and which models they use. At Ascenzia, agents are created in Ascenzia Studio, then Ascenzia Sentinel steers them and keeps the log of their executions. The platform is hosted in France, in the Paris region, or in Switzerland with our partner Hidora. Where the platform is hosted is not where the models run: that depends on the models you approve.

Shadow AI means public AI tools that teams use without the company having decided it. Ascenzia Sentinel shows which public AI tools your teams use and blocks what you decide to block: the platform applies your rules, it does not invent rules for you.

By giving each agent a capped budget in euros, a catalogue of approved models, guardrails and an exportable execution log. That is what Ascenzia Sentinel does for the agents on the platform. One limit: the control tower only sees agents on the platform, not those another tool runs elsewhere.

Ascenzia Studio is where you create your agents: you describe the task in plain language and the agent is created and connected to your information system. Ascenzia Sentinel is where you steer them: approved models, guardrails, a capped budget per agent, costs in euros and the execution log. Agents created in Ascenzia Studio are attached to Ascenzia Sentinel from the start.

In France, in the Paris region, for our French clients; in Switzerland, with our partner Hidora, for our Swiss clients. On our side, only metadata comes back (the status and duration of a call), never the content of your prompts or of the answers.

Yes. Your agents talk to the multi-model gateway in Ascenzia Sentinel, which sends each request to the model you have approved. Apart from a fallback when a model goes down, the model never changes by itself: you decide when to change it.

The subscription to Ascenzia Studio and Ascenzia Sentinel is €390, €990 or €2,900 excl. VAT per month depending on the plan (small business, SME, mid-cap), on request for a multi-site group, with a six-month commitment. The plan depends on usage, not on company size. Agent consumption is billed separately, at cost and with no margin. A custom-built agent is covered by a monthly fee per agent, quoted on request.

It gives you evidence to show, not turnkey compliance: the execution log and its exports, costs in euros, approved models and guardrails. The obligations remain yours, and no platform fulfils them for you. Your DPO or legal counsel still has their own analysis to do.

Not necessarily. If you are unsure about your first use cases, we offer a process audit to spot where AI saves time, then training for your teams; the platform comes after. A 20-minute call is enough to scope a first agent.

Twenty minutes to scope your first agent.

YOUR AI AGENTS. STEERED. SECURED. SOVEREIGN.