GitHub

Free

Make your AI apps safe.

FreeFree tier

Inputs: text

Type

Open Source

Company

superagent-ai

Visit Website GitHub (40 stars)

About GitHub

Superagent SDK is an open-source SDK for AI agent safety, designed to protect AI applications against prompt injections, data leaks, and harmful outputs. It provides runtime guard capabilities to detect and block malicious instructions, redact PII/PHI/secrets from text, scan repositories for AI-targeted attacks like repo poisoning, and run red team scenarios against production agents (test feature coming soon). The SDK works with any LLM provider (OpenAI, Anthropic, Google, Groq, Bedrock, etc.) and offers low-latency guard inference using open-weight models that can run on your own infrastructure. Integration options include TypeScript and Python SDKs, a CLI tool, and an MCP server for use with Claude Code and Claude Desktop. Licensed under MIT.

Key Features

Guard: Detect and block prompt injections, malicious instructions, and unsafe tool calls at runtime

Redact: Automatically remove PII, PHI, and secrets from text using LLMs

Scan: Analyze repositories for AI agent-targeted attacks such as repo poisoning and malicious instructions

Test: Run red team scenarios against production agents (coming soon)

Works with any LLM provider (OpenAI, Anthropic, Google, Groq, Bedrock, etc.)

Open-weight models for self-hosted guard inference with 50-100ms latency

Low latency optimized for runtime use

MIT license with full transparency

Integration options: TypeScript SDK, Python SDK, CLI, MCP Server

Supports both cloud API and self-hosted deployments

Pros & Cons

Pros

Open-source (MIT) with full transparency
Low latency guard inference (50-100ms) suitable for real-time applications
Works with any LLM provider, avoiding vendor lock-in
Supports self-hosted deployment of guard models for data privacy
Multiple integration options (SDKs, CLI, MCP) for flexible adoption
Covers multiple safety dimensions: injection blocking, redaction, repo scanning, red teaming
Active development with community support via Discord and GitHub

Cons

Test (red team) feature is currently listed as 'coming soon' and not yet available
Requires API key from superagent.sh for cloud features, which may introduce dependency on external service
Relatively new project with potentially limited community resources compared to established solutions

Best For

Runtime safety for AI chatbots and agentsPreventing prompt injection attacks in productionRedacting sensitive information (PII, PHI, secrets) from user inputsSecurity auditing of code repositories for AI-specific threatsCompliance and proving safety measures to customersRed team testing AI agent endpoints for vulnerabilities

FAQ

What is Superagent SDK?

Superagent SDK is an open-source toolkit for AI agent safety that helps block prompt injections, redact sensitive data, scan repositories for threats, and run red team scenarios. It provides SDKs for TypeScript and Python, a CLI, and an MCP server.

What models does Superagent SDK support?

Superagent SDK works with any LLM provider, including OpenAI, Anthropic, Google, Groq, Bedrock, and others. It also supports open-weight models that you can run on your own infrastructure.

Is Superagent SDK open-source?

Yes, Superagent SDK is released under the MIT license, making it fully open-source with transparent code.

How do I get started with Superagent SDK?

You can get started by signing up at superagent.sh to obtain an API key, then install the SDK via npm (npm install safety-agent) or pip (uv add safety-agent), and set your API key as an environment variable.

Can I run Superagent SDK on my own infrastructure?

Yes, the Guard feature supports open-weight models that can be deployed on your own infrastructure for low-latency inference (50-100ms), providing data privacy and control.