AgentGuard provides local guardrails for AI coding agents, offering DLP, dependency protection, token optimization, and observability.

How does AgentGuard work?

AgentGuard acts as a local boundary for AI agent requests, applying Guardian or Saver controls to prompts, tool inputs, and file reads before they reach the model.

Is AgentGuard free to use?

Yes, AgentGuard is offered for free.

What operating systems are supported by AgentGuard?

AgentGuard supports macOS and Linux operating systems.

What kind of token savings can I expect with AgentGuard?

AgentGuard compresses shell output, caches repeated file reads, and shrinks noisy responses to reduce token consumption.

How does AgentGuard protect against compromised libraries?

AgentGuard can block agents from using compromised libraries, protecting against attacks or agent mistakes.

AgentGuard

Guardian Agent, Token Savings for Claude Code and Codex

Agents

FREE

In a single click, with 0 complicated setup Save costs (tokens) with Claude Code: - smartly reference cache for already read files - compress shell outputs - better replacements for grep, find etc Guardian Agent for coding agents: - Smart disallow for sensitive data reading by...

Features

Token Savings
Uses smart caching to reduce token usage for Claude when dealing with code.
Shell Compression
Makes shell outputs smaller, saving on token costs.
Smarter Tools
Offers alternatives to tools like grep and find, optimized for cost and performance.
Sensitive Data Protection
Stops coding agents from accessing sensitive information.
Library Updates
Encourages agents to use the latest library versions.
Compromised Library Block
Prevents agents from using unsafe or hacked libraries.
Attack Prevention
Protects against attacks and mistakes made by the agent.
Agent Safety
Acts like a guard to keep your coding agents secure and efficient.