Usage & Analytics

Token Usage & Cost Savings

Agent-CoreX reduces the number of tool schemas sent to your LLM on each request. The Usage dashboard shows an estimated dollar value of the tokens you've saved this month.

Last updated: April 2026

Where to find token data

Navigate to Dashboard → Usage. Scroll to the bottom of the page — below the 30-day bar chart and the quota progress bar — to find the "Estimated cost savings this month" card.

Cost savings card

The card displays two pieces of information:

Estimated cost savings this month

Based on avg 60% token reduction across N queries.

$X.XX

saved vs. unrouted

•$X.XX — the estimated dollar saving for the current 30-day period
•N queries — the total query count in the last 30 days, same value shown in the "Queries this period" stat card

How it is calculated

The savings figure uses a fixed formula:

estimated_savings = total_queries_last_30_days × $0.07

The $0.07 per query figure represents the average token cost saved when Agent-CoreX routes only the relevant 5% of tools instead of your full tool schema, assuming:

•Average 60% reduction in tokens sent to the LLM per request
•Typical LLM input pricing across GPT-4, Claude, and Gemini models

This is an estimate based on averages. Actual savings depend on your LLM provider, model pricing, and the size of your full tool schema. Use it as a directional indicator rather than an exact billing figure.

Why tokens matter

Most LLM providers charge per input token. When you attach 100 tool schemas to every agent request — even if the agent only uses 3 of them — you pay for all 100 schemas on every call. With a large tool registry, this can represent 40–70% of your total token spend.

Agent-CoreX intercepts that step: your agent sends a natural-language query, gets back only the top 5 (or however many you configure with top_k) relevant tools, and sends only those to the LLM. See How Tool Retrieval Works for details on the ranking algorithm.