Superagent

Superagent is a web platform for enterprise security, engineering, and compliance teams to red team test production AI agents and validate safety controls before issues reach users.

Key capabilities include:

• Black-box red team testing that deploys specialized attack agents against live systems to uncover data leaks, harmful outputs, and unwanted or unauthorized actions, delivered with findings, evidence, and remediation guidance
• Runtime defenses with reasoning-driven threat detection for prompt injections, jailbreaks, and malicious tool calls, plus input/output filtering to protect secrets, PII, and internal context
• Tool and code execution safeguards designed to block unsafe automations, privilege escalation, and hidden jailbreak instructions in files or attachments
• Unified observability with live traces, audit trails, and policy controls, plus a shareable Safety Page for procurement and security reviews

Pricing: Partly open source (its SDK and tooling), but its hosted API and production features are paid on a usage-based model.

Notable for combining adversarial red teaming with deployable guardrails and an active open-source ecosystem, including the Superagent SDK (TypeScript and Python) compatible with OpenAI, Anthropic, and Google, plus additional safety projects such as Brin and open-weight guardrail models.