Aegis
Local-first defense across the entire AI attack surface.
Prompt injection, AI-generated phishing, data exfiltration, malware — modern threats hit every layer. Aegis is 15 integrated components covering browser, mobile, IDE, email, API, endpoint and malware, all analysing on-device. No cloud dependency. No telemetry by default.
0 / 15
components online
0
prompt injections blocked
0
injection patterns · 9 classes
0+
YARA rules · 100% EICAR
0
telemetry sent
<5ms
injection scan latency
Representative dashboard — not a live instance.
●Prompt-injection prevention — 200+ patterns across 9 threat classes (role hijack, instruction override, jailbreak, data-exfil, encoding bypass, delimiter injection…) with tunable sensitivity.
●15 integrated shields — network, file-trust, kernel, LLM defense, browser, mobile VPN, API-gateway WAF, IDE secret-scanning, email/phishing, threat-intel, and enterprise fleet management.
●Malware & rootkit detection — YARA engine with static, behavioral, memory and rootkit analysis; guaranteed EICAR detection.
●Embeddable SDKs — JavaScript and Python, zero external dependencies, three-line integration into any app.
●Enterprise-ready — RBAC, policy templates, and CEF / LEEF / STIX export for fleet-wide correlation, all air-gap capable.
0+
automated tests passing
0
integrated components
0
prompt-injection patterns
100%
EICAR malware detection
RustTypeScript / ReactPythonSwift · KotlinTauri 2Axum · TokioYARASQLite WAL
ℹ
Work in progress — reference implementation. Every component is real, compiling, tested code, but the suite isn't shipped: the kernel driver is unsigned (a Windows code-signing certificate and a dedicated test machine are needed to validate on real kernels), and the SDKs, extensions and apps aren't published. Production-grade code, not yet a released product.
🔒
Architecture and verified results shown. Source, detection rules and binaries are private — this is a capability showcase, not a distribution.
Securing an
AI product?
I build defense that runs where your users are and knows exactly how AI systems get attacked. Let's talk about your threat model.
Start a project →