8. Exact vs Estimated vs Hybrid: Token Cost Tracking in a Multi-Agent FinWise Pipeline

1. INTRODUCTION

In Articles 1 through 7, I talked about and built architecture, routing, messaging, persistence, tooling, and client UX. Now we address one of the most practical questions in production and business:

How many tokens did this request actually use across all agents?

In a multi-agent system, this is not trivial. A single user message can involve:

• Supervisor classification,
• one or more downstream specialist calls,
• mixed availability of exact provider usage.

FinWise solves this using a unified token model with three modes:

• exact
• estimated
• hybrid

This article explains the token measurement structure, how supervisor/downstream/total are populated, and how the MAUI app shows the right tally for each mode.

2. USEFUL TOOLS, RESOURCES AND CODEBASES USED

Below are the key technologies and resources used in this article: