At Bravent, we have built a functional demo — Citadel Token Monitor — that demonstrates how organizations can control the usage of AI agents without modifying a single line of their code.
This is what we learned while building it, and why we believe that AI agent governance will become one of the most important foundations for organizations looking to scale artificial intelligence securely, efficiently, and sustainably.
The Question Almost Nobody Is Asking
Over the last few months, companies have moved from:
“We want to do something with Generative AI”
to:
“We already have ten AI agents running in production.”
Entire departments — legal, HR, sales, and customer service — are now relying on intelligent assistants capable of reading internal documents, answering questions, and automating critical business processes.
And then the uncomfortable question appears:
Who is actually controlling all of this?
- How many AI agents do we really have deployed?
- How much does each one consume every month?
- What happens if an agent suddenly drives token consumption through the roof?
- How do we prevent agents from accessing sensitive information they shouldn’t see?
- How can we demonstrate governance and traceability to auditors, security teams, or finance leaders?
The reality is that most organizations still don’t have clear answers.
Not because they don’t care, but because the AI governance layer is not yet part of the standard architecture of many AI initiatives.
Organizations build the agent, connect it to the model, deploy it into production… and conversations about cost control, quotas, security, and governance are postponed until later.
The Technical Problem Explained in Business Terms
Every AI agent consumes tokens.
And every token costs money.
The challenge is that token consumption is usually measured at the model level, not at the application, team, or department level.
The result:
It’s like having ten departments sharing a single corporate credit card without individual statements.
- Impossible to allocate costs accurately
- Impossible to audit
- Difficult to control
Impossible to scale confidently
What We Built: Citadel Token Monitor
To demonstrate that this challenge can be solved today using existing Azure technologies, Bravent’s Innovation Team developed Citadel Token Monitor.
The demo simulates a fictional construction company with three specialized agents:
- Legal
- Human Resources
- Sales
Each agent has access only to its own internal knowledge base:
- Corporate bylaws
- Collective bargaining agreements
- Real estate sales catalogues
But the most important component isn’t the agents themselves.
The real value lies in the real-time governance layer built on top of them.
What the Demo Monitors in Real Time
✅ Individual Token Tracking per Agent
Every request is measured and attributed to the correct agent.
✅ Department-Level Monthly Quotas
For example:
- Legal → 200,000 tokens
- HR → 300,000 tokens
- Sales → 500,000 tokens
The limits are fully configurable according to business requirements.
✅ Automatic Consumption Limits
If an agent exceeds its assigned quota:
- the system blocks further requests,
- informs users about the waiting period,
- and can redirect traffic to another available agent.
✅ Cost Estimation and Monthly Forecasting
The platform calculates in real time:
- accumulated consumption,
- current spend,
- and projected month-end costs.
✅ Content Safety and Risk Controls
Every interaction passes through safety controls designed to prevent harmful or inappropriate responses.
✅ Observability and Auditability
All usage metrics are stored in Application Insights, providing:
- audit trails,
- operational visibility,
- historical analysis,
- and usage pattern monitoring.
Most importantly, all of this happens without changing the code of any AI agent.
Governance policies live in the gateway, not inside each individual application.
The Architecture in Four Sentences
The core component is Azure API Management (APIM) acting as an AI Gateway, following the architecture pattern Microsoft is promoting through Foundry Citadel.
Each agent receives its own identity through subscription keys, and every request passes through the gateway before reaching the AI model.
APIM applies native AI policies such as:
- azure-openai-token-limit
- azure-openai-emit-token-metric
- Azure Content Safety
These policies enable organizations to:
- track token consumption,
- enforce quotas,
- filter content,
- and monitor usage in real time.
On top of that, we added:
- a Router Agent built with Azure AI Foundry,
- a real-time React dashboard,
- and an Azure Functions backend responsible for protecting agent credentials.
The Result
Organizations can give AI agents greater autonomy without sacrificing visibility, governance, or cost control.
Why This Matters Right Now
Organizations deploying AI agents in 2026 will increasingly fall into two categories:
Those that invested early in AI governance
and
Those that eventually face:
- unexpected AI bills,
- security incidents,
- or governance challenges they cannot explain during audits.
AI agent governance is no longer a future concern.
It is a conversation happening right now among:
- CIOs
- CISOs
- CFOs
- Innovation teams
- Platform and architecture leaders
How Bravent Can Help
At Bravent, we have spent years helping organizations deploy enterprise-grade AI solutions on Azure.
What we consistently see is the same pattern:
- Prototypes move fast.
- Production deployments slow down as soon as governance, security, compliance, and cost management enter the discussion.
That’s exactly why we built this demo.
To demonstrate, in just a few minutes, how these challenges can be solved using proven technologies and architectures.
We can help you:
- Audit your current AI landscape and identify governance blind spots
- Design and implement an AI Gateway based on Azure APIM
- Define agent identities, quotas, and governance policies
- Integrate observability, monitoring, and auditing capabilities
- Implement content filtering and alerting mechanisms
- Evolve toward secure multi-agent architectures
Is your organization already deploying AI agents?
Now is the right time to build a governance layer before growth becomes difficult to control.
📩 Talk to the Bravent team
Contact us at: info@bravent.net and we’ll show you how to implement an AI governance framework tailored to your organization, your cost model, and your security requirements.
