AgentGuard is a solution for AI coding agents, providing local guardrails to ensure secure and efficient coding practices.

Overview

AgentGuard offers a simple flow and local boundary, allowing users to guard every prompt and spend fewer tokens. It blocks secrets, risky installs, and oversized requests, while also compressing shell, search, diff, and JSON noise.

Key Features

• Guardian Agent: changes behavior before the request leaves, with DLP on prompts, tool inputs, and file reads, token caps for oversized requests, and vulnerability guard on installs and dependency writes.
• Token Saver: cuts waste before it reaches the model, with compression of verbose shell output, caching of repeated file reads, and shrinking of noisy search, diff, and JSON responses.
• Flow: simple flow and local boundary, with requests going in, Guardian or Saver acting, and review surfaces keeping the trail visible.

How It Works

The workflow is straightforward: requests go in, Guardian or Saver acts, and review surfaces keep the trail visible. Users can review the result in Logs, Analytics, and Garden.

Benefits

The benefits of using AgentGuard include increased security, reduced token usage, and improved efficiency.

Use Cases

AgentGuard can be used with various AI coding agents, including Claude Code, Codex CLI, and Cursor.

Integrations

AgentGuard supports integrations with Claude Code, Codex CLI, and Cursor, with different depths of coverage.

Pricing Overview

AgentGuard is free and open-source, with no pricing details available.