Cerbere-AG is a three-layer security guardrail for AI agents: Regex, ML, and an LLM Judge. AI agents are unpredictable. They improvise and act with full autonomy. I built Cerbere-AG (named after the three-headed dog, Cerberus) so that autonomy doesn’t come at the cost of control. Why three layers? Our benchmarks show it: → Regex alone lets more than 30% of attacks through. → ML pushes recall up to 100% in some of our tests, but raises false positives. → The LLM Judge resolves the ambiguous cases ML can’t classify. It also looks at context. Cerbere-AG follows tool calls, checks whether they’re legitimate, and then allows them, blocks them, or flags them for real-time human approval. Agents keep their autonomy, and humans keep control. Everything is audited too. The collector gathers spans and traces into an audit trail that is independent from the agent and supports compliance. And like any tool, it has limits. That’s why observability matters: security isn’t only about blocking attacks, it’s about understanding what happened, what was allowed, what was blocked, and why. Open source and free for now. Try it and give me your honest feedback 👇 github.com/chrismsmr-celcom/… app.cerbereag.site

Prompt filters check what your agent is told. They can't stop it from leaking your .env.Cerbère intercepts every LLM & tool call before execution: taint tracking (catches encoded exfiltration), budgets, human approval. Open source 👇 github.com/chrismsmr-celcom/…

Be among the first to build, deploy, and scale AI agents safely without losing control. Join us in building the future of AI agent security. Get started with Cerbere-AG, an open-source security and observability layer designed to help teams deploy AI agents with confidence. 100% free. $0. No credit card required. Let’s build together. app.cerbereag.site

Your AI agent just tried to send sensitive data through an external tool. Would your team even know? With Cerbere-AG, you can see what agents are doing, what gets blocked, why it was blocked, and where the risks are emerging. Runtime security, enforcement, and observability in one place. Here's what that looks like in practice ↓

Giving an AI agent access to your APIs, internal tools, and sensitive data is a security decision — not just an engineering one. Cerbere-AG is building the security infrastructure that lets companies and engineering teams deploy AI agents with greater control and confidence. Monitor agent execution. Track untrusted data. Enforce security policies. Audit tool calls. The goal: know what your agents are doing, understand where risks emerge, and control what they’re allowed to do. Deploy AI agents without giving up control. 🔗 github.com/chrismsmr-celcom/…

🚀 Introducing Taint Tracking in Cerbere-AG. AI agents don’t just process prompts. They read webpages, retrieve documents, and pass data between tools. What happens when untrusted data influences a sensitive action? Cerbere-AG’s new taint tracking capability follows untrusted data through agent execution, helping identify risky data flows before they reach sensitive operations. Track the data. Trace its path. Understand the risk. Built for safer AI agent execution. 🔗 github.com/chrismsmr-celcom/…

We’ve just added two new features to Cerbere AG: >Shadow Mode >Incident Mode Both are designed to give developers more visibility and control over their AI agents. Launch demo video coming tomorrow. Open-source. Free. Built for AI agent security. github.com/chrismsmr-celcom/…

Prompt injection defense isn't about catching bad words. It's about watching what untrusted data DOES. Here an agent exfiltrates a secret encoded in base64. No keywords. Blocked anyway.

I started without a plan: an API key, a Supabase account, and the desire to build something agentic. An observability runtime, then security on top of it, and then the real question: who controls what the agents do? 855 commits later, it's called Cerbère AG. Self-taught, brick by brick. Runtime security for AI agents. Open-core, Apache 2.0. github.com/chrismsmr-celcom/…

My choice: B for production, potentially with LangGraph inside Temporal activities for complex agent reasoning. I’d also add Cerbere AG as the runtime security and control layer to enforce policies, secure tool calls, manage human approvals, and maintain audit trails.

My work in chrismsmr-celcom/cerbere-AG was selected for the XERJ Corpus Hub: AI coding agents retrieve it as a worked reference instead of guessing. hub.xerj.org/contributors/ch… Thank you @d0znpp @CerbereAg

Cerbere AG was designed to make the deployment of AI agents safer and more autonomous without you losing control. Watch the demo video on YouTube:youtu.be/TfyxTCkgXqA?si=IlmQ…

Why Cerbere? Vigilance. Three heads, three layers: fast rules, ML, and an LLM judge. And one principle: nothing reaches the real world without being authorized. Autonomy needs boundaries. Cerbere enforces them at runtime.

AI agents are moving into production, but the damage is becoming harder to ignore. A 2026 Cloud Security Alliance survey found that 65% of organizations surveyed experienced an AI agent-related incident in the past 12 months. I’m building Cerbere AG to help prevent that: runtime security, control, and observability for AI agents. github.com/chrismsmr-celcom/…

I ran Cerbere AG against 2,003 test cases: 1,000 attacks, 703 benign prompts, and 300 hard negatives. Results with regex + ML: • 100% attack recall • 571.8 ms p95 latency • 2.28% false-positive rate on benign prompts • 1.33% false-positive rate on hard negatives The trade-off is clear: catching attacks is only half the problem. Legitimate requests must keep working, too. The next challenge is reducing false positives without losing detection coverage. Building runtime security for AI agents, one benchmark at a time.