TokenTracker: AI Cost Calculator Agent

A helper agent that estimates the cost of running AI agents, tools, and servers — before production.

Installation

• Create & Activate Python venv in root directory
• pip install -r ./requirements.txt
• Create .env in my_agent/
• Add to .env the following values

GOOGLE_GENAI_USE_VERTEXAI=0
GOOGLE_API_KEY=<your-api-key>

• Run adk run my_agent in root directory for cli version
• Run adk web --port <port> in root directory for web version

Agent Purpose & Scope

Background & Motivation:

I wanted to build an AI Agent for real-world use, but I realized I had no clear way to estimate how much it would cost a user — especially if multiple agents, tools, hosting platforms, or LLM models were involved. Because of this uncertainty, I had an idea:

➡️ Build an AI Agent that can calculate the cost of running any AI agent or full AI project. This tool could help me — and other developers — plan projects, control budget, optimize models, and compare platforms before launching to production.

Agent Purpose

The Cost Calculator AI Agent is designed to:

• Estimate the cost of AI models.
• Calculate server and hosting costs.
• Simulate multi-agent workflows and their total cost
• Compare pricing options across LLM models and platforms.
• Help developers predict production cost before deployment.
• Support decision making: Recommend cheaper or optimized alternative.

Core Functions:

1. Collect information

The agent needs to know what the AI is buid of before being able to calculate the cost of production:

• Detect missing inputs: “We can’t calculate cost yet — data is missing.”
• Offer defaults: “Would you like industry averages?”
• Ask smart questions : “Do you expect few or many tasks per day?”
• Show the formula: Build trust & transparency
• Refine over time : Make estimates more accurate

2. calculate production costs

• LLM Cost Estimation: Calculates cost based on tokens & model pricing
• Server Cost Estimation: Estimates hosting cost for websites or agents
• Tool Cost Estimation: Costs of APIs (e.g., web search, vector DB)
• Multi-Agent Workflow Cost: Simulates team of agents with multiple steps
• Cost Comparison: Compare platforms/models to find cheapest option
• Recommendation: Suggest cheaper alternatives or cost optimizations

Scope & Limitations

1. Core Features :

• LLM Cost Estimation : Calculates token-based cost for GPT / Claude models
• Server Cost Estimation: Estimates monthly cost for hosting providers
• Tool Cost Estimation: Cost of APIs (e.g., web search, vector DB)
• Multi-Agent Simulation: Calculates cost of workflows with multiple agents
• Cost Comparison : Compares platforms and model providers
• Recommendations : Optimizes for lowest-cost setups

2. Future Expansion

In later versions, the agent could:

• Use real pricing APIs (AWS Pricing API, OpenAI tools API, etc.)
• Track real token usage of agents
• Store history and learn typical usage patterns
• Run as a service with a web interface

Architecture Overview:

Multi-Agent Flow:

1️⃣ Overview of the Workflow

User Input:

User specifies project details:

Number/type of AI agents

Expected API calls per agent (tokens, calls per day)

Hosting requirements (Vertex AI resources, storage, traffic)

Whether live pricing is needed

Planner Agent :

• Interprets the input
• Decides which tasks are needed (LLM cost, server cost, live pricing)
• Generates a task list for the Dispatcher

Dispatcher Agent:

Sends each task to the appropriate agent:

• LLM Cost Agent → calculates token usage cost based on model choice
• Server Cost Agent → calculates Vertex AI resource cost
• Pricing API Agent → fetches live pricing using Google API key
• Workflow Agent → aggregates multiple agents’ costs

Pricing API Agent:

Uses Google Cloud APIs (or Vertex AI pricing endpoints) to fetch:

• Vertex AI managed instance costs
• Storage costs (Cloud Storage buckets)
• Traffic / egress fees
• LLM pricing if using Vertex-hosted models
• Stores results in JSON format for other agents

LLM Cost Agent:

Uses static or live model pricing

• Calculates total tokens per month × cost per 1M tokens

Server Cost Agent:

Calculates Vertex AI compute costs:

• Instance type (CPU/GPU) × hours per month × region pricing
• Storage and network usage

Workflow Agent:

Aggregates all costs:

• Multi-agent LLM usage
• Hosting + storage + bandwidth

Formatter Agent:

Creates human-readable report and JSON output:

✅ Result:

• Understand the project
• Collect pricing data (live + static)
• Compute LLM + server + workflow costs
• Output clear, actionable reports

Answer always includes:

• 🧾 Source of prices

• 📅 Date of retrieval

• 📊 Confidence level

User Input → Planner Agent │ ▼ Dispatcher Agent ┌─────────┬─────────┐ ▼ ▼ ▼ LLM Cost Server Cost Pricing API Agent Agent Agent │ │ │ └─────────┴─────────┘ │ ▼ Workflow Agent │ ▼ Formatter Agent → Output