▎ 👋 Hey Product Hunt! ▎ ▎ I'm Yash, the maker of SoterAI. ▎ ▎ I kept seeing teams ship AI chatbots, RAG apps, and agents to ▎ production with basically zero security — one clever prompt could ▎ leak customer data, jailbreak the system, or make an agent call ▎ tools it never should. The existing tools were either dumb keyword ▎ filters (too many false alarms) or observability dashboards that ▎ only tell you after something broke. ▎ ▎ So I built SoterAI — a real-time security command layer that sits ▎ between your app and your LLM. ▎ ▎ What it does: ▎ 🛡️ Blocks prompt injection, jailbreaks & data/PII leaks before the ▎ model runs ▎ 🤖 Agent Firewall — stops rogue tool calls & risky multi-step ▎ actions in real-time ▎ 🔍 32 security services across 6 layers (Monitor, Protect, Detect, ▎ Control, Compliance, Manage) ▎ ⚡ <50ms latency — fast enough for live chat ▎ 🇮🇳 India-first PII detection (Aadhaar, PAN, UPI, GSTIN…) ▎ ▎ The part I'm proud of: on our adversarial benchmark it caught 97/97 ▎ attacks with 0% false positives (F1 = 1.0) — most security tools ▎ force you to choose between catching attacks and annoying real ▎ users. This doesn't. ▎ ▎ It's open-core — self-host with Docker, or use the hosted version. ▎ SDKs for JS, Python, LangChain, LlamaIndex & more, so you're ▎ protected in ~5 lines of code. ▎ ▎ 🎮 Try the live playground (throw your nastiest prompts at it): ▎ https://soterai.publicvm.com/pla... ▎ 📖 Docs: https://soterai.publicvm.com/docs ▎ ⭐ GitHub: https://github.com/yashchauhan66... ▎ ▎ I'd genuinely love your feedback — especially what would stop you ▎ from trusting an AI security layer in production? I'll be here all ▎ day answering everything. 🙏

SoterAI

The firewall for AI apps and agents — 0% false positives

The firewall for AI apps and agents — 0% false positives