cyb3r

cyb3r is an offensive-security reasoning model project for a Pi-based agent harness.

The model is one part of a larger operator stack:

• Pi agent provides the runtime layer; in the system analogy, think of it as the operating-system layer.
• cyb3r is the agent harness that coordinates planning, execution, reassessment, and tool use.
• The model provides penetration-testing reasoning, branch ranking, pivot decisions, and operator discipline; think of it as the CPU inside the harness.
• pi-pen-kit is the tool layer for reverse shells, Sliver, Metasploit, and operator-level C2 workflows.

Current Status

Training is in progress. We are currently in the first epoch of an epoch 01-03 reasoning-focused pass.

The current corpus is a cybersecurity-focused training corpus of approximately 1.6B cybersecurity-related training rows. The first training focus is reasoning behavior:

• structured penetration-testing reasoning
• global reassessment during multi-step tasks
• attack-path branch ranking
• reduced rabbit-hole fixation
• preserved tool-call competence

After the first reasoning-focused pass, we plan to broaden and strengthen security knowledge across tools, techniques, procedures, vulnerabilities, and defensive context.

Hardware / Partner Note

Training hardware is currently the limiting factor. We are actively looking for partners who can help with compute, evaluation capacity, applied security feedback, or controlled real-world workflows.

Intended Use

cyb3r is intended for authorized security research, lab environments, defensive validation, and controlled penetration-testing workflows.

It is not intended for unauthorized access, abuse, or deployment against systems without permission.

Availability

This repository is a teaser/model-card placeholder while the first training run is in progress. Weights, adapters, evaluation notes, and usage examples will be added as artifacts become available.

Related Tooling

pi-pen-kit provides Pi extensions for offensive-security workflows:

• controlled reverse-shell listeners and session handling
• Sliver C2 operator workflows
• Metasploit RPC workflows
• interactive shell read/write/interrupt/probe/exec primitives