DEV Community: Christopher dikesa

I’m building a free, open-source security layer for AI agents — and I want the community to help build it

Christopher dikesa — Sun, 04 Oct 2026 19:59:23 +0000

I’m building Cerbere AG, an open-source security and observability runtime for AI agents.
The idea is simple:
See what your agent does in production. Detect risky behavior. Block dangerous actions. Flag ambiguous ones. And require human approval when the system shouldn’t make the decision alone.
But this post isn’t really about announcing another AI security product.
It’s about making this a community project.
I want to make Cerbere AG free for everyone.
No credit card.
No paid plan required to start.
No artificial wall preventing developers from experimenting with agent security.
If you’re building an agent with tools, APIs, MCP, browser automation, databases, or other real-world capabilities, you should be able to add a security layer without first having to convince your company to spend thousands of dollars.
And I’m not pretending it’s finished.
It isn’t.
There are probably things I’ve missed.
There are edge cases I haven’t thought about.
There are attacks I haven’t covered.
There are integrations that could be better.
There are probably architectural decisions that someone in this community will look at and say:
“You should do this differently.”
Good. Tell me.
If you find a bug, report it.
If you find a security issue, report it.
If you think the detection logic is wrong, challenge it.
If you have a better approach, open a PR.
If you want to add an integration, build it.
If you have a new attack scenario, add it to the benchmark.
If you use Cerbere and discover that something doesn’t work, tell me.
The goal isn’t for me to personally build every feature.
The goal is to build something useful enough that other people want to contribute to it.
I’m building this largely by myself from Kinshasa, DRC, and I don’t have a large engineering team behind me.
So I can’t realistically cover every framework, every agent architecture, every tool integration, and every attack pattern alone.
That’s exactly why I’m opening it up.
What Cerbere currently focuses on
Agent execution tracing
Tool-call monitoring
Risk detection
Blocking
Flagging
Human approval for ambiguous/sensitive actions
Audit logs
Security evidence
Agent behavior observability
The project is still evolving, and I would rather have people tell me where it is wrong than pretend it is already perfect.
If you’re building AI agents, try it.
Use it.
Break it.
Audit it.
Contribute to it.
Tell me what sucks.
And if you think the idea is useful, give the repository a ⭐ or share it with someone building agents.
The more people test it, the more useful it can become.
Free. Open source. No credit card. Built in public.
GitHub: https://github.com/chrismsmr-celcom/cerbere-AG
If you see a problem, don’t just complain about it.
Bring the problem to the repo. Let’s build the solution together.

I’m building Cerbère-AG, a security evidence layer for AI agents.

Christopher dikesa — Sat, 26 Sep 2026 19:27:56 +0000

I’m building Cerbère-AG, a security evidence layer for AI agents.

Most AI security tools focus on what goes into the model: prompt injection, malicious inputs, jailbreaks, etc.

I’m focusing on what happens after the model decides to act.

Cerbère-AG observes and controls agent actions across tool calls, including:

tool-call monitoring and traces
policy enforcement
sensitive-action detection
argument and capability checks
budgets and execution limits
trajectory-level risk detection
human approval for sensitive actions
security evidence for AI agent activity

The idea is simple:

Don’t just ask whether an agent is safe to talk to. Ask whether it is safe to let it act.

I’m looking for developers and teams running AI agents in real or realistic environments to test Cerbère-AG and tell me where it fails.

I’m especially interested in design partners who can give real-world feedback on agent workflows, policies, approvals, and failure cases.

If you build AI agents, security tooling, MCP integrations, or autonomous workflows:

→ Give me your feedback: what would you expect a production-grade agent security layer to catch that Cerbère currently doesn't?

And if you find the project useful, a ⭐ on GitHub helps people discover it.

I’m more interested in breaking it and finding its weaknesses than in compliments.

If you have an agent that you think could expose a real failure mode, send it my way.

→ Test it:
→ GitHub:

We benchmarked our prompt-injection detector against OWASP's LLM Top 10

Christopher dikesa — Sat, 26 Sep 2026 19:11:17 +0000

We just published our first public benchmark for AgentGuard, the runtime we're building to secure AI agents in production.
The honest version: our deterministic (regex) layer alone catches 91.5% of prompt injection attempts with zero false positives, in under 2ms. Adding a ML layer pushes recall to 98.1% — but the trade-offs are real: ~450ms latency, and a 33% false-positive rate on benign prompts that were deliberately worded to look like attacks.
We're publishing the numbers as they are, weaknesses included, and mapped everything to the OWASP Top 10 for LLM Applications so it's easy to compare.

DEV Community: Christopher dikesa

I’m building a free, open-source security layer for AI agents — and I want the community to help build it

I’m building **Cerbère-AG**, a security evidence layer for AI agents.

We benchmarked our prompt-injection detector against OWASP's LLM Top 10

I’m building Cerbère-AG, a security evidence layer for AI agents.