wcoreiron/regent-aegis-core

Regent-Aegis Core

Regent-Aegis Core is a Phi-3 3.8B model fine-tuned for structured defensive cybersecurity decisions.

Paste security data — CVE alerts, scan output, log snippets, endpoint findings — and get a prioritized decision with severity, MITRE ATT&CK mappings, recommended actions, verification steps, and rollback guidance.

Try it live

**⚔️ Open in Regent-Aegis Space →**

Model Details

Output Contract

The model outputs a structured JSON object with:

{
  "mode": "triage | advisory | remediation | incident",
  "severity": "critical | high | medium | low | informational",
  "confidence": 0.0,
  "zero_trust_pillars": ["identity", "device", ...],
  "evidence_required": ["..."],
  "analysis": "...",
  "recommended_actions": [
    {"action": "...", "risk": "...", "how": "...", "scope": "..."}
  ],
  "verification": ["..."],
  "rollback": ["..."],
  "mitre_attack": ["T1190", ...],
  "d3fend": ["D3-DA0001", ...],
  "regent_policy_ids": ["..."]
}

Use Cases

• —Vulnerability triage and remediation prioritization
• —Incident response planning
• —Identity and access governance review
• —Endpoint and browser posture interpretation
• —Cloud misconfiguration analysis
• —Defensive security workflow assistance

Intended Use

Defensive security workflows only.

Not for offensive exploitation, evasion, credential theft, persistence, or unauthorized access.

Limitations

• —May produce incorrect security conclusions — always verify with current threat intel
• —CVE exploitability changes; use CISA KEV and EPSS for live decisions
• —Q4KM quantization trades minor quality for memory efficiency vs F16

License

Code: MIT. Model weights derived from Phi-3-family — downstream users must comply with Microsoft Phi-3 license terms.