Launching today
AegisKernel
Zero-Trust AI Agent Security Runtime
1 follower
Zero-Trust AI Agent Security Runtime
1 follower
Experience why probabilistic LLM guardrails fail and how deterministic capability boundaries block 100% of autonomous tool hijacking in sub-millisecond time.
Hey Product Hunt!
I'm Adi, a 16-year-old solo systems engineer. I built AegisKernel because autonomous AI coding agents (CrewAI, LangChain, Claude Code) are getting terminal access, but prompt guardrails completely fail at the shell boundary.
Once an agent executes bash, an indirect prompt injection or poisoned diff can run rm -rf, spawn reverse shells, or leak .env and SSH keys. Secondary LLM evaluators are too slow and probabilistic to stop this.
🛡️ How AegisKernel works:
Instead of asking another LLM "does this prompt look safe?", AegisKernel enforces security deterministically at the OS syscall boundary:
- Sub-0.016ms Latency: Enforces path-scoped capability tokens in-process with zero LLM token cost.
- Deterministic Interception: Terminates unauthorized execve, sensitive path tampering, and shell breakouts before execution.
- Open Ecosystem: Works with LangChain, CrewAI, AutoGen, and connects with Aegisora for full-stack policy governance.
- 100% Free & Open Source on PyPI (`pip install aegis-kernel`) and GitHub.
🧪 Links:
- Live Web Sandbox: https://adikshatriyag-crypto.github.io/aegis-kernel/
- GitHub: https://github.com/adikshatriyag-crypto/aegis-kernel
Currently onboarding 2–3 design partners building autonomous or coding agents to test our 10-line staging wrapper.
Would love your honest feedback, critique, and thoughts on how your team handles agent bash security!