agent-guard: A Drop-in Spend Cap and Local Audit Log for LLM API Calls

Overview

agent-guard is a lightweight tool designed to stop AI agents from spending too much money on API calls. It was created after a real incident where a misconfigured agent loop on Cloudflare Durable Objects generated a $34,000 bill in just eight days because there were no real-time safeguards. This tool wraps your LLM SDK calls to enforce a hard dollar ceiling and keep a local record of every action taken.

Benefits

The main advantage of agent-guard is that it prevents runaway costs by stopping API calls once a set budget is reached. It works entirely on your local machine, so you do not need to set up external services or dashboards. The tool is privacy-focused because it only stores a hash of the prompt by default, meaning the actual text of your prompts is not logged unless you choose to enable that feature. It also requires no account to use, making it easy to integrate into any project.

Use Cases

This tool is ideal for developers who want to build AI agents without worrying about unexpected bills. It works with popular providers like OpenAI and Anthropic. You can set the budget to reset every day, for each session, or globally across all restarts. The tool is also useful for teams that need to audit their AI usage locally without sending data to the cloud. It can be installed in your project directory and used to wrap existing SDK methods with just a few lines of code.

Pricing

agent-guard is an open-source tool released under the MIT license. It is free to use and does not require any subscription or account setup. The tool itself does not charge for usage, but it helps you manage the costs of the AI APIs you connect to it.

Vibes

Developers appreciate agent-guard for its simplicity and reliability. It solves a critical problem of cost control in AI development without adding complexity to the workflow. Users value the fact that it runs locally and does not send sensitive data to external servers. The tool has gained attention for its ability to prevent financial disasters caused by misconfigured AI agents.

Additional Information

agent-guard is built for Node.js and requires version 18 or higher, with version 22.5 preferred for native SQLite support. It is available as an npm package and can be installed using standard package managers like npm, pnpm, or yarn. The tool includes a command-line interface for monitoring spend and managing logs. It does not have built-in adapters for frameworks like LangChain, so users must wrap raw SDK methods directly. The project is actively maintained and updated to support new features and providers.