High-Fidelity Architecture
A comprehensive technical deep-dive into the sovereign engine: 3-Layer Separation of Concerns, Axum/Tokio Async Execution, Sapphire Shield Governance, and LanceDB GraphRAG Memory.
Sovereign Protocol Core Stack
Click any component below to inspect runtime specs, data flows, and code contracts.
Directives SOP
Layer 1: Instruction Set
The entry point of the sovereign engine. Directives are authored in Markdown and YAML, ensuring business logic is human-readable and machine-executable.
struct Directive {
id: Uuid,
sop_version: String,
parameters: HashMap<String, Value>,
constraints: Vec<Constraint>,
} Operational Perspectives
Select a role to see how AI-Tadpole-OS transforms security, governance, operational control, and daily execution.
Air-Gapped Local Security & Merkle Audits
IT departments define execution boundaries. When configured for air-gapped operation, the Hard Privacy Gate can disable external network access while tool activity is recorded for audit review.
Continuous Telemetry & Human-In-The-Loop
Managers gain visibility over multi-agent workflows with
10Hz MessagePack telemetry streams and WebSocket
HITL gates. High-risk capabilities such as
shell:execute and budget:spend require
management approval.
Sovereignty as the Ultimate Enterprise Advantage
Deploying Tadpole OS on organization-controlled hardware can reduce API dependency and keep proprietary context within the security boundaries selected for that deployment.
Transition from Manual Drudgery to Swarm Director
Operators turn routine SOPs into executable Markdown directives
in directives/, supporting scaled work while
maintaining human oversight over execution outcomes.
Ingest playbooks, coordinate parallel reviews, and synthesize findings with operator oversight.
Tri-Slot Model Routing Architecture
AI-Tadpole-OS decouples model selection from application logic. The engine can route prompts across three fallback slots to improve resilience when a provider or model is unavailable.
Local Ollama Engine
Local qwen2.5-coder / llama3.3 models running air-gapped on hardware GPU/NPU acceleration.
Groq Ultra-Speed LPU
Ultra-fast cloud inference fallback (500+ tokens/sec) for complex multi-agent reasoning tasks.
Google Gemini 1.5 Flash
Massive 2M token context window slot for large document ingestion & deep architectural synthesis.
Core Protocol Performance Metrics
Representative runtime targets for the Rust server-rs
and Tauri desktop builds.
Ready to Deploy Sovereign Swarms?
Clone the runtime repository and start building autonomous, privacy-first multi-agent systems on your own hardware.