On-Demand: Controlling AI Costs: Optimize and Chargeback LLM Usage with agentgateway
As AI agents multiply across enterprise environments, LLM usage visibility and cost control become mission-critical. Traditional monitoring tools struggle to track token consumption, attribute costs, and enforce accountability across multi-agent, multi-org workflows.
In this webinar, you'll get a deep technical look at how agentgateway provides gateway-level observability to track usage in real time, calculate costs, enforce budgets, and implement chargeback models across teams.
What you’ll learn:
- A breakdown of LLM consumption challenges in agent-based systems and why token tracking + attribution require more than legacy monitoring
- How to configure agentgateway for usage monitoring: budgets, rate limits, and multi-provider routing (Anthropic, OpenAI, xAI, and more)
- Practical chargeback and reporting techniques for per-user/per-team cost breakdowns and billing system integrations
- A live walkthrough of the agentgateway-llm-consumption-demo, simulating multi-agent usage and visualizing cost flows in Prometheus and Jaeger
Frequently asked questions
How does agentgateway track LLM token usage?
agentgateway proxies requests to LLM providers such as OpenAI, Anthropic, and xAI, so it sees every request and response and can record input and output token counts per user, team, agent, and model. It exports this data as metrics and traces, for example to Prometheus and Jaeger, for dashboards and cost reports.
What is AI chargeback?
AI chargeback is the practice of attributing LLM and AI infrastructure costs to the teams, products, or customers that generated them, then billing or reporting those costs back. It depends on accurate per-request attribution, which is why it works best when measured at a gateway that sees all AI traffic.
How do I enforce LLM budgets at the gateway?
Configure token-based rate limits and usage budgets on agentgateway per user, team, or API key. Requests that exceed a limit are rejected or rerouted at the gateway. Because enforcement happens in the data plane, it applies consistently across every agent and LLM provider without changes to application code.
As AI agents multiply across enterprise environments, LLM usage visibility and cost control become mission-critical. Traditional monitoring tools struggle to track token consumption, attribute costs, and enforce accountability across multi-agent, multi-org workflows.
In this webinar, you'll get a deep technical look at how agentgateway provides gateway-level observability to track usage in real time, calculate costs, enforce budgets, and implement chargeback models across teams.
What you’ll learn:
- A breakdown of LLM consumption challenges in agent-based systems and why token tracking + attribution require more than legacy monitoring
- How to configure agentgateway for usage monitoring: budgets, rate limits, and multi-provider routing (Anthropic, OpenAI, xAI, and more)
- Practical chargeback and reporting techniques for per-user/per-team cost breakdowns and billing system integrations
- A live walkthrough of the agentgateway-llm-consumption-demo, simulating multi-agent usage and visualizing cost flows in Prometheus and Jaeger
Frequently asked questions
How does agentgateway track LLM token usage?
agentgateway proxies requests to LLM providers such as OpenAI, Anthropic, and xAI, so it sees every request and response and can record input and output token counts per user, team, agent, and model. It exports this data as metrics and traces, for example to Prometheus and Jaeger, for dashboards and cost reports.
What is AI chargeback?
AI chargeback is the practice of attributing LLM and AI infrastructure costs to the teams, products, or customers that generated them, then billing or reporting those costs back. It depends on accurate per-request attribution, which is why it works best when measured at a gateway that sees all AI traffic.
How do I enforce LLM budgets at the gateway?
Configure token-based rate limits and usage budgets on agentgateway per user, team, or API key. Requests that exceed a limit are rejected or rerouted at the gateway. Because enforcement happens in the data plane, it applies consistently across every agent and LLM provider without changes to application code.
Speakers
Sebastian Maniak
Sebastian Maniak is an AI Architect at Solo.io, where he focuses on go-to-market for AI connectivity — agent gateways, LLM security, and the infrastructure that makes agentic systems deliverable, secure, and operable in the enterprise. He is also a professor of network and cybersecurity at Sheridan College, teaching Applied Network Security and Infrastructure as Code. He brings more than two decades of experience in application delivery, security, and automation — from F5, Fortinet, and HashiCorp through cloud-native service mesh and API platforms — and now applies that same delivery-and-control mindset to Enterprise AI. Over his career he has worked across architecture, consulting, product, education, and content, which gives him a broad generalist’s view of how technical platforms actually get adopted.
Michael Levan
Focusing on providing business value through technology, working directly with customers, providing first-class advisory, and content, Michael Levan helps organizations address needs through Agentic AI practices. He has worked across industries as an engineer, consultant, trainer, and content creator. He's a CNCF Ambassador, AAIF Ambassador, 4x published author, international speaker, and was part of the Kubernetes v1.28 and v1.31 Release Teams. Michael's approach translates technical complexity into business value, delivering actionable solutions that enhance performance, security, observability, and agentic workloads.
