superagent

Superagent SDK is an open-source safety layer for AI applications, offering four core functions: Guard (runtime detection and blocking of prompt injections and unsafe tool calls), Redact (automatic removal of PII, PHI, and secrets from text, replacing items like email addresses and SSNs with labeled placeholders), Scan (analysis of GitHub repositories for agent-targeted attacks such as repo poisoning), and Test (upcoming red-team scenario runner for production agents). It integrates with Claude specifically through an MCP server compatible with both Claude Code and Claude Desktop, while TypeScript and Python SDKs allow direct embedding into any application. The library works across Anthropic, OpenAI, Google, and other model providers. A notable self-hosting option exists via three open-weight Guard models ranging from 0.6B to 4B parameters, available in GGUF format for CPU inference, allowing organizations to run threat detection on their own infrastructure without sending data to external APIs. Security engineers and compliance teams building Claude-powered products are the primary audience.

Lo que la gente pregunta sobre superagent