Cerbere Ag

The three-headed agent guardrail for AI agents.
Complete security & observability runtime.
Secure. Observe. Automate. 🛡️

[Cerbere Ag @CerbereAg]

@lightsilver323 https://t.co/h2K09YYds0

[Christopher Dikesa @dikesa_dev]

We’re looking for a few early design partners.

If you’re building or deploying AI agents in production — or you’re a startup building agent-based products — I’d like to hear from you.

We’re building Cerbere-AG, a runtime security and observability layer for AI agents.

We’re looking for teams willing to:
→ integrate Cerbere-AG with an agent
→ test it in a real workflow
→ tell us what works and what doesn’t
→ challenge our security controls with real use cases

No polished sales demo. We’re looking for honest technical feedback that can shape the product.

If you’re deploying AI agents and want to participate, reply here or DM me.

[Christopher Dikesa @dikesa_dev]

https://t.co/vPGkZstkuO

[Cerbere Ag @CerbereAg]

@dikesa_dev 🚀🚀🚀

[Christopher Dikesa @dikesa_dev]

An agent, a tool, a single execution limit.

Each call is verified, recorded and attributable.

No agent goes unnoticed.

We connected Cerbère-AG on a real AI agent in production.

The SDK is located between the agent and its tools - no modelling change is required.

(What we learned in prod: when the collector is unreachable, the default must be fail-closed, not fail-open.

A guardrail that lets everything through when it's down isn't a guardrail)

Github ⭐️: https://t.co/PJpYuMZPr3
Try cerbereag: https://t.co/H9PpnkOzjc

[Christopher Dikesa @dikesa_dev]

Your AI agent is in production. What happens after you give it access?

Files. Tools. APIs. Data. Tokens. Actions.

Cerbère-AG (@CerbereAg) is a runtime layer to observe, detect, enforce and audit what your agents do — before the tool call fires.

3 layers at the execution boundary:
1. Rules / Regex
2. ML classifier
3. LLM judge
→ ALLOW / BLOCK

No model changes required. No agent passes unseen.

🐕 https://t.co/H9PpnkOzjc
⭐️ https://t.co/PJpYuMZPr3

[Christopher Dikesa @dikesa_dev]

Your AI agent is in production. What happens after you give it access?

Files. Tools. APIs. Data. Tokens. Actions.

Cerbere-AG gives you a runtime layer to observe, detect, enforce and audit what your agents do in production.

No card required. Your data stays home.

Try it → https://t.co/H9PpnkO1tE
pip install cerbereag

[Christopher Dikesa @dikesa_dev]

🐕‍🦺 CerbereAG — Setup Guide

Runtime security for AI agents · Latest version: cerbere-ag 0.4.2

1. Install the Python SDK

pip install cerbere-ag>=0.4.2

2. Get your API key

Create an account on https://t.co/SI5UyOErod → dashboard → generate your AGENTGUARD_API_KEY, then:

export AGENTGUARD_API_KEY="your-key-here"

3. Protect your agent (SDK)

Every prompt and tool call is inspected by the policy engine before execution.

from cerbere_ag import AgentGuard

guard = AgentGuard(api_key="your-key-here")

@guard.protect
def run_agent(prompt):
...

4. Connect your MCP agent (Claude Code, Cursor, Cline…)

Code-free alternative: the CerbereAG MCP server.

pip install cerbere-ag-mcp>=0.4.2
claude mcp add cerbereag --transport sse https://t.co/zgMbK4SFlf

5. Verify in the dashboard

Go to https://t.co/SI5UyOErod → Recent Events / Audit Trail:

Event

Decision

tool_call: read_file

✅ ALLOWED

tool_call: email exfiltration

❌ BLOCKED

tool_call: dangerous command

❌ BLOCKED

What CerbereAG 0.4.2 gives you

Real-time inspection of prompts and tool calls (Python SDK, MCP, HTTP gateway)

Blocking: data exfiltration, dangerous commands, indirect prompt injections

Centralized audit trail — every decision logged, compliance-ready

Compatibility: Claude · Cursor · Cline · Anthropic API agents · LangGraph · CrewAI · Composio

[Cerbere Ag @CerbereAg]

@dikesa_dev 🚀🚀🚀

[Cerbere Ag @CerbereAg]

RT @dikesa_dev: Exfiltration of secrets, RCE via a simple file read by the agent. It happened to Cursor (CVE-2025-59944), to Devin, and the…

[Cerbere Ag @CerbereAg]

@Sarakhan49309 https://t.co/h2K09YXFCs

[Cerbere Ag @CerbereAg]

@BuildByAman Hey

[Christopher Dikesa @dikesa_dev]

Today, trying to exfiltrate a file through an AI agent by simply saying “ignore your instructions” is becoming less effective. Agents can have filters and regex rules that detect obvious attacks.

But the real problem is deeper: an agent may not be able to distinguish between a legitimate action and a dangerous one.

That’s why I created Cerbere-AG @CerbereAg.

Cerbere-AG is a security and observability runtime for companies deploying AI agents in production and sensitive environments.

If an agent tries to move a sensitive document from one folder or server to another, the action can be detected, evaluated, and require approval before execution.

No approval, no action.

The human remains in control.

[Cerbere Ag @CerbereAg]

@FlippedRay https://t.co/S9DhvpyAft

[Cerbere Ag @CerbereAg]

@Sarakhan49309 https://t.co/S9DhvpyAft

[Cerbere Ag @CerbereAg]

@BuildByAman Let’s connect !!

[Cerbere Ag @CerbereAg]

@BuildByAman https://t.co/h2K09YXFCs

[Cerbere Ag @CerbereAg]

@Sarakhan49309 We make autonomous AI agents safe and observable so it doesn’t cause you any harm

https://t.co/h2K09YXFCs

[Cerbere Ag @CerbereAg]

@GauravGoyalAI @X Hey

[Cerbere Ag @CerbereAg]

@GauravGoyalAI Let’s connect