Journal · Guides
How to work out what you are spending.
Written to be useful whether or not you use Capsera. Two of these recommend a different tool where a different tool is the right answer.
Agent cost is a measurement problem before it is a spending problem. Almost every team that thinks it has a budgeting question actually has an attribution one: the invoice arrives as a single total, nothing in it maps to an agent, a customer or a run, and so there is nothing to make a decision against. These guides are about closing that gap, whichever tool you close it with.
Start with the evaluation checklist if you are comparing tools, or with finding your agents if you are not yet sure how many you are running. Most teams cannot answer that second question, and it is the one everything else depends on.
- How to evaluate an LLM cost toolNine questions that separate call-level observability from agent-level governance, with honest answers for Langfuse, LiteLLM and Capsera — including the four where another tool wins.9 questions
- Find the agents already in your codebaseMost teams cannot say how many agents they are running. This is how to count them, and how to tell which ones make live model calls.Diagnostic
- Gateway vs observability vs governanceThree layers that get sold as the same thing and solve different problems. Which one you need depends on the unit you have to control.Comparison
- Every question we have been askedThe full researched Q&A on agent cost attribution, budgets and enforcement, answered in one place.Q&A hub
- GlossaryAgentic spend, cost defect, per-run attribution and the rest of the vocabulary, defined once so the rest of the site can use it.Definitions