In development

Secros, an AI assistant for security analysis.

Secros is being built to help with defensive security work. It isn't publicly available yet, and this page explains what it will do and where it stands today.

What it's for

Help for defenders, not attackers.

How it works

A planner and specialist agents.

The prototype plans each request, then hands it to one of four specialist agents: security analysis, SOC and incident work, research, or coding. The research agent can read public sources; the coding agent works as a small team that writes, tests and reviews code.

Until Secracy's own model is ready, Secros runs on an existing open-weight model.

Status
Assistant prototype Working internally
Safety policy Written, in testing
Public access Not yet
SecrosLM model Research, not yet trained
Safety policy

What Secros will and won't do.

Will help with
  • Defensive analysis and hardening
  • Incident response
  • Vulnerability triage
  • Explaining malware behaviour at a safe level
  • Tutoring on legal CTF challenges and labs
Will refuse
  • Exploiting real targets
  • Credential theft and phishing
  • Creating malware, persistence or evasion
  • Scanning systems you aren't authorised to test

Deeper technical explanation is reserved for requests clearly framed as a legal lab, a sandbox, a challenge, or a system you own.

Research

SecrosLM

SecrosLM is Secracy's research into a custom model architecture for multi-step security reasoning. Rather than stacking many unique layers, it runs one shared block of layers in a loop, so it can spend more computation on harder problems. It uses a mixture of experts so different parts of the network can specialise.

The architecture is written but the model has not been trained yet.

Stay in touch

Want to hear when it's ready?

Email us and we'll let you know when Secros opens up.

Email contact@secracy.in
Credits

Who's building it.

Secros is built inside Secracy by a small team. The work spans the assistant itself, the agent orchestration behind it, and the SecrosLM architecture research.