local-agent-cost-tracker

Real-time cost tracking for AI agents—know exactly what you're spending before the bill arrives.

What is this?

A lightweight desktop application that monitors your AI agent traffic in real-time, calculates token costs across OpenAI, Anthropic, and local models, and alerts you before you exceed your budget. Built to solve the #1 pain point from production agent deployments: "I don't know how much this is costing me until the invoice arrives." Runs entirely offline with zero telemetry, backed by SQLite, and complements observability tools by focusing purely on cost tracking.

Features

• Real-time cost tracking – Monitor token consumption and costs as they happen
• Multi-model support – OpenAI, Anthropic, and local LLM pricing built-in
• Budget alerts – Get notified before you exceed spending thresholds
• Offline-first – Zero telemetry, no cloud dependencies, runs on your machine
• SQLite backend – Simple, fast, portable cost history
• CLI + Server – Command-line tool or HTTP server for integration
• Docker ready – Includes Dockerfile and docker-compose for easy deployment
• Zero dependencies on external services – Works with air-gapped deployments

Quick Start

Installation

# Clone and install
git clone <repo-url>
cd local-agent-cost-tracker
pip install -e .

Configuration

Copy .env.example to .env and configure your settings:

cp .env.example .env

Key variables:

• OPENAI_API_KEY – For OpenAI model pricing (optional)
• ANTHROPIC_API_KEY – For Anthropic model pricing (optional)
• BUDGET_LIMIT – Monthly spending cap in USD
• ALERT_THRESHOLD – Alert when spending reaches X% of budget

Run the Server

# Start the tracking server
python -m local_agent_cost_tracker.server

# Server runs at http://localhost:8000

Command-line Usage

# Track a single API call
local-agent-cost-tracker log --model gpt-4 --tokens 150 --cost 0.0045

# View cost summary
local-agent-cost-tracker summary --period month

# Export cost history
local-agent-cost-tracker export --format csv --output costs.csv

# Check remaining budget
local-agent-cost-tracker budget

Docker

docker-compose up -d

Usage Example

from local_agent_cost_tracker import CostTracker

tracker = CostTracker()

# Log a request
tracker.log_request(
    model="gpt-4",
    input_tokens=120,
    output_tokens=85,
    timestamp="2025-03-29T10:30:00Z"
)

# Get cost summary
summary = tracker.get_summary(period="week")
print(f"Weekly spend: ${summary['total_cost']:.2f}")
print(f"Budget remaining: ${summary['remaining_budget']:.2f}")

# Check if over budget
if tracker.is_over_budget():
    print("⚠️ Budget exceeded!")

Tech Stack

• Language: Python 3.10+
• HTTP Server: FastAPI
• Database: SQLite
• Pricing Engine: Custom model pricing resolver
• Token Counting: tiktoken (OpenAI), built-in counters
• Testing: pytest
• Deployment: Docker, docker-compose
• CI/CD: GitHub Actions

License

MIT – See LICENSE for details.

Get started tracking costs in 5 minutes. No signup, no telemetry, no surprises.