llmtrack

LLM cost attribution per feature — track which parts of your product are spending your AI budget.

The Problem

Every company using LLMs knows their total monthly bill, but almost none know which user-facing feature or internal workflow is responsible. Without granular attribution, teams cannot optimize high-cost features, set feature-level unit economics, or enforce sub-budgets.

Installation

pip install llm-cost-track

Quickstart

import openai
from llmtrack import CostTracker

tracker = CostTracker()

with tracker.feature("document_summary"):
    response = openai.chat.completions.create(
        model="gpt-4o",
        messages=[{"role": "user", "content": "Summarize this text: ..."}]
    )

tracker.report()

Features

• 🎯 Feature-Level Cost Attribution: Tag LLM calls with clear product or feature names.
• ⚡ Zero-Config Auto-Patching: Automatically captures usage from OpenAI and Anthropic SDKs with safe fallbacks.
• 🧵 Thread-Safe Context Manager: Safely isolate concurrent requests and nested execution flows.
• 💰 Built-in Model Pricing: Accurate per-million token pricing for 25+ major models across OpenAI, Anthropic, Google, Mistral, Meta, and DeepSeek.
• 📊 Rich Terminal & HTML Reports: Generate rich ASCII terminal breakdown tables or self-contained dark-mode HTML dashboards.
• 🚨 Budget Alerts: Set daily spending limits per feature with custom webhook/callback notifications.
• 💾 Lightweight & Local Storage: Defaults to embedded SQLite (llmtrack.db) or in-memory storage.
• 💻 CLI Tooling: Query stats and manage tracking databases directly from the terminal with llmtrack.

Supported Models

Custom and unknown models fallback gracefully or can be dynamically registered with register_custom_model().

API Reference

CostTracker

tracker = CostTracker(
    storage: Optional[BaseStorage] = None,
    db_path: str = "llmtrack.db",
    auto_patch: bool = True
)

• with tracker.feature(name: str):
  Context manager tagging all calls in the block with name.
• tracker.log_call(model: str, input_tokens: int, output_tokens: int, feature: Optional[str] = None, latency_ms: float = 0.0, metadata: dict = {}) -> CallEvent
  Manually record an LLM call event.
• tracker.report(days: int = 7, output: str = "terminal", filepath: Optional[str] = None) -> None
  Display or export a cost attribution report (output="terminal" or output="html").
• tracker.summary(days: int = 7) -> dict
  Return an aggregated summary dict with total cost, calls, and per-feature breakdowns.
• tracker.set_budget_alert(feature: str, daily_limit_usd: float, callback: Optional[Callable] = None) -> None
  Register daily spending alerts.

CLI Usage

# View terminal summary table (default: last 7 days)
llmtrack report

# Generate report for the last 30 days
llmtrack report --days 30

# Export self-contained HTML report
llmtrack report --html --output cost_report.html

# Clear database
llmtrack clear --db llmtrack.db

Storage Backends

• SQLiteStorage(db_path="llmtrack.db"): Default persistent storage. Auto-creates schema and indexes.
• MemoryStorage(): Thread-safe in-memory store, ideal for testing, serverless functions, or short-lived scripts.

from llmtrack import CostTracker, MemoryStorage

tracker = CostTracker(storage=MemoryStorage(), auto_patch=False)

Contributing

Contributions, issues, and feature requests are welcome! Feel free to check the issues page.

1. Fork the repository
2. Create your feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.