Superagent

Infrastructure for coding agents. Builders of VibeKit.

The AI firewall — stop attacks, leaks, and model abuse

Superagent builds infrastructure for running and embedding AI coding agents in real workflows — safely, flexibly, and with full control. Our latest open-source release is VibeKit, a CLI wrapper for coding agents like Claude Code, Gemini CLI, and Codex. It runs agent-generated code in isolated Docker containers, redacts secrets automatically, and streams output in real-time. No cloud signup required — works locally by default. VibeKit lets developers swap models behind any agent CLI, see exactly what agents do (not just text output), and run agents without affecting their local environment. Teams use it to meet compliance requirements while giving developers the tools they actually want to use. Learn more: https://www.vibekit.sh/

Superagent is the world’s first firewall purpose-built for AI systems. It protects your models and data at runtime by inspecting every prompt and response. It blocks prompt injections, jailbreaks, and data leaks, prevents backdoors in generated code, and quarantines suspicious outputs. The firewall is powered by a small language model fine-tuned for security and safety. Unlike static rules or regex filters, it can reason about context to catch subtle attacks and leaks. The model is lightweight and optimized for speed — fast enough to run inline with every request and response without slowing down your apps. Superagent also provides observability with full traces, audit logs, and policy enforcement, giving teams security and control without slowing down AI adoption. Learn more: https://www.superagent.sh/