🛡️ TokenFence Examples — Cost Circuit Breaker for AI Agents

Stop runaway AI costs in 2 lines of code. Budget caps, automatic model downgrade, and kill switches for OpenAI, Anthropic, and Google Gemini.

The Problem

AI agents in production are a cost ticking bomb:

• GPT-4o: $10/1M output tokens
• Claude Opus: $75/1M output tokens
• Average agent run: 15-50 API calls
• One bad loop: 💸💸💸 surprise $200+ bill

A single runaway subagent loop can burn through your entire monthly budget in minutes. Rate limits don't help — they cap requests, not dollars.

The Solution

from tokenfence import guard
import openai

# That's it. $5 budget cap, auto-downgrade at 80%, kill at 100%.
client = guard(openai.OpenAI(), budget=5.00)

# Use normally — TokenFence tracks every token
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Analyze this data..."}]
)

TokenFence wraps your existing OpenAI/Anthropic client with zero code changes:

• Budget caps — hard spending limits per workflow, per agent, per task
• Auto model downgrade — when spend hits 80%, automatically switch from GPT-4o → GPT-4o-mini
• Kill switch — hard stop when budget is exhausted, returns safe error instead of more spending
• Cost tracking — real-time spend visibility across all your agents

Examples

Example	What It Shows	Provider	Difficulty
basic-budget	Simple per-workflow budget with kill switch	OpenAI	⭐ Beginner
auto-downgrade	Automatic model downgrade when budget runs low	OpenAI	⭐ Beginner
anthropic-guard	Protect Anthropic Claude API calls	Anthropic	⭐ Beginner
multi-agent	Budget tracking across a multi-agent workflow	OpenAI	⭐⭐ Intermediate

Documentation

• 📖 Quickstart Guide — Get running in 5 minutes
• 📚 API Reference — Full API docs with all parameters
• 💰 Model Pricing Table — 40+ models with current per-token costs

Installation

Python

pip install tokenfence

# With provider extras
pip install tokenfence[openai]        # OpenAI support
pip install tokenfence[anthropic]     # Anthropic support
pip install tokenfence[all]           # All providers

Node.js / TypeScript

npm install tokenfence

import { guard } from 'tokenfence';
import OpenAI from 'openai';

const client = guard(new OpenAI(), { budget: 5.00 });

Why Not Just Use Rate Limits?

Feature	Rate Limits	TokenFence
Caps requests per minute	✅	—
Caps dollar spend	❌	✅
Per-workflow budgets	❌	✅
Auto model downgrade	❌	✅
Kill switch on overspend	❌	✅
Works across providers	❌	✅
Real-time cost tracking	❌	✅

Rate limits protect the API provider. TokenFence protects your wallet.

Supported Providers & Models

• OpenAI: GPT-4o, GPT-4o-mini, GPT-4-turbo, GPT-3.5-turbo, o1, o1-mini, o3-mini, GPT-5.4 mini/nano
• Anthropic: Claude Opus 4, Sonnet 4, Claude 3.7 Sonnet, Claude 3.5 Sonnet/Haiku, Claude 3 Opus/Sonnet/Haiku
• Google: Gemini 2.5 Pro/Flash, Gemini 2.0 Flash, Gemini 1.5 Pro/Flash
• DeepSeek: DeepSeek Chat, DeepSeek Reasoner

Use Cases

• 🤖 Multi-agent orchestration — Give each agent its own budget envelope
• 🔄 Automated pipelines — Prevent cost blowups in CI/CD AI workflows
• 📊 Cost allocation — Track spend per feature, per user, per team
• 🛡️ Production safety — Hard kill switch prevents billing surprises
• 📉 Smart degradation — Automatically use cheaper models when budget is tight

Links

• 🌐 TokenFence Website
• 📦 Python SDK on PyPI
• 📦 Node.js SDK on npm
• 📝 Blog Posts — 57 articles on AI cost control
• 📖 Blog Source

License

MIT — use these examples however you like.