Superagent — the AI Firewall 🥷🔥
Stops prompt injections, backdoors, and data leaks.
Hi everyone, we’re Alan and Ismail, founders of Superagent (YC W24).
We built the world’s first AI Firewall — powered by a model that sits between you and your models, protecting every request and response in real time.
The Problem
AI is quickly becoming the foundation of how software is built and used. But every time you send a prompt or receive a response, you take a risk:
- Models can be tricked with prompt injections or jailbreaks
- Sensitive data like API keys and PII can leak out
- Malicious code or backdoors can slip through responses
If you’re building with AI or adopting third-party AI tools, you’re exposed.
Our Solution
Superagent introduces NinjaLM — a small language model fine-tuned for security and safety. It runs in runtime (tens of milliseconds) and reasons about every prompt and response before it reaches your system.
✅ Stops prompt injections and jailbreaks
✅ Prevents secret leaks (API keys, credentials, PII)
✅ Blocks malicious code before it reaches your system
✅ Full audit logs, traces, and observability for compliance
From internal apps to tools like Claude Code or ChatGPT, Superagent protects your AI without slowing you down.
Backstory
Before Superagent, we built plenty of AI apps ourselves. Each time, we ran into the same problem: adding even the most basic kind of protection was incredibly hard.
Traditional firewalls are built for static rules — not reasoning. But AI isn’t static. It thinks, it adapts, and that creates a completely new class of security challenges.
We realized the only way forward was to fight fire with fire: build a model that could reason about other models, catching attacks, leaks, and malicious behavior in real time. That’s how the AI Firewall was born.
Try It
We just launched our hosted service, and the open-source repo is live. Engineers can get started in minutes, and executives get the auditability and compliance they need.
👉 superagent.sh
👉 github.com/superagent-ai/superagent
Call to Action
If you’re building or adopting AI:
- Engineers — run your apps through Superagent today
- Security teams — add prevention and compliance controls
- Partners — we’re exploring integrations and distribution
We’d love your feedback. Links in comments.
