Sovereign AI Infrastructure
Identity Secured.
Keys Protected.
Zero Exposure.
Aieges is a deployable security layer for your AI runtime. Secure your Vertex, Anthropic, and OpenAI keys with Ledger-based identity verification and MFA. Inspect every tool call. Audit every action.
From RJS Software Labs + DuoNode Inc
Sovereign AI goes further
Runtime is half the story. Security is the other half.
Omnipotent puts your AI team on your hardware. But what about the calls those agents make? What about the prompts they receive? That's where Aieges comes in.
Act II — Aieges
Every AI Tool Call. Inspected.
Your AI agents call external tools — APIs, databases, code runners. The gateway doesn't block your workflow. Aho-Corasick pattern matching runs inline at <2ms. Everything else is inspected after it passes through — a breadcrumb-logging agent traces every call, and when a trail is marked bad, it blocks that pattern from coming back.
API CallINSPECTEDDB QueryLOGGEDCode ExecPOLICY-CHECKEDFile WriteBLOCKEDLayer 1 — Inline
Aho-Corasick engine · 147+ threat patterns · <2ms per request
Layer 2 — Async Inference
Llama inference model · Breadcrumb trail analysis · Blocks bad actors retroactively
No unauthorized data exfiltration
No unreviewed tool calls
Full audit trail
Aieges Sentinel
AI Security That Never Phones Home
Prompt injection detection, PII leak prevention, malicious payload scanning — all running on your device. Not in some vendor's cloud. Your data never leaves your machine.
147+
Threat Patterns
0
Cloud Dependency
100%
Offline Capable
Pattern-matching engine with 147+ threat signatures. Regex-based prompt injection detection, PII pattern scanning, and payload blocklists. Zero external dependencies.
147+ threat patternsRegex + blocklist engineSub-millisecond latencyZero dependenciesEverything in Shield, plus an AI classification layer. Transformer-based intent analysis catches novel attacks that bypass static patterns. Still runs on your device.
All Shield protectionsTransformer intent classifierNovel attack detectionLocal inference onlyFull AI security stack. Rules + classifier + a local LLM for deep contextual reasoning about threats. Catches sophisticated multi-turn prompt injection and subtle data exfiltration attempts.
All Developer protectionsLocal LLM reasoning engineMulti-turn attack analysisDeep context evaluationOpen source. Free forever for personal use.
Act I
Omnipotent — Your AI Team, Your Hardware
Your agents answer to a SaaS. Shouldn't they answer to you? Omnipotent is a self-hosted agentic runtime with 12 specialized AI agents, three-tier persistent memory, and self-correcting protocols. Local-first, no cloud dependency.
12
Agents
3
Memory Tiers
0
Cloud Dependencies
4
Model Support
Twelve specialized AI agents — Architect, Forgemaster, Oracle, Sentinel, and eight more — running on your hardware. No SaaS. No API middlemen. Full sovereignty.
Self-correcting Prophecy protocolConsensus-gated destructive opsMulti-model hot-swap (Claude, Gemini, Codex, Grok)The Blackboard + Prophecy pipeline breaks strategic goals into executable atoms. Planning, execution, and review in a closed loop — no human micromanagement required.
Prophecy strategic planningBlackboard shared workspaceAirlock data isolation protocolHot memory for active context. Warm memory for session history. Cold memory for persistent knowledge. All stored locally — your data never leaves your machine.
Hot: Active context windowWarm: Session graph (Graphiti)Cold: Filesystem checkpoint