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
| Property | Value |
|---|---|
| Base model | Phi-3 3.8B |
| Parameters | 3.8B |
| Context length | 4096 |
| Fine-tuning | LoRA v4.1 (defensive security decisions) |
| Quantization | Q4_K_M (~2.2 GB) |
| Format | GGUF |
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
- Q4_K_M 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.