superagent
Protect AI applications against prompt injections, data leaks, and harmful outputs.
Block prompt injections, redact PII, and scan agent repositories for threats.
Superagent is an open-source SDK designed to secure AI agent applications at runtime. It targets compliance and security professionals who need to enforce guardrails and prove safety to customers.
What it is
The SDK provides runtime tools to detect prompt injections, block unsafe tool calls, and automatically redact PII, PHI, and secrets from text using models like OpenAI or open-weight alternatives. It also scans code repositories for agent-targeted attacks and supports integration via TypeScript, Python, a CLI, and an MCP server.
Who it's for
It suits AI engineers and compliance officers building production LLM apps who need verifiable guardrails. It does not suit teams looking for a no-code security dashboard or those not writing TypeScript or Python applications.
Key features
Runtime Guardrails
Detect and block prompt injections, malicious instructions, and unsafe tool calls.
Automated Redaction
Remove PII, PHI, and secrets from text automatically using configurable models.
Repository Scanning
Analyze code repositories for AI agent-targeted attacks and repo poisoning.
Red Team Scenarios
Run automated red team scenarios against production agents to discover vulnerabilities.
Up and running in 3 steps
Our honest verdict
What's great
- Supports multiple languages including TypeScript and Python SDKs
- Compatible with any major model provider or open-weight models
- Includes specialized MCP server integration for Claude Code and Desktop
- Provides automated PII and secret redaction out of the box
Where it falls short
- Requires a paid API key from superagent.sh to use the managed service
- Red team testing functionality is marked as coming soon
- Relies on external AI models which incur ongoing API usage costs
Pricing & license
The SDK is free under its MIT license, but you must sign up at superagent.sh to get an API key, which may involve hosted service costs.
Open remote AI-training roles
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
What models does Superagent support?
It works with OpenAI, Anthropic, Google, Groq, Bedrock, and open-weight models.
Can I run the guard locally?
Yes, you can run open-weight guard models on your own infrastructure with low latency.
How do I integrate Superagent into my workflow?
You can use the TypeScript or Python SDKs, the CLI tool, or the MCP server for Claude.
