Sovereign Intelligence

PurpleDirective

An AI operations system built for a single human operator. One COO agent — Violet — receives every task, triages it, delegates it, and synthesizes the result. The system learns from every session.

Meet Violet

Violet is the COO. She is not a chatbot — she is an operations system with a hierarchy beneath her.

Human (final authority — you) │ └── Violet (COO — receives tasks, triages, delegates, audits, synthesizes) │ ├── E.I.K. Advisory Council (3 agents, 2-round deliberation, hard stop) │ ├── Evolution — forward motion, innovation, "what should we build?" │ ├── Improvement — verification, accuracy, "is this actually right?" │ └── Keenness — blind spots, edge cases, "what are we not seeing?" │ └── Operational Staff (domain execution agents) ├── Domain agents scaffolded from a shared template └── Each with: directives, tools, identity, memory

How a Task Moves Through the System

Task Arrives

Human sends a request. Violet receives it and immediately classifies it — simple, complex, or administrative.

↓

Triage

Simple — routed directly to the relevant domain agent with context and expected output format.
Complex — escalated to the E.I.K. Advisory Council.
Admin — handled by Violet directly.

↓

Deliberation Complex tasks only

Round 1: All three advisory agents analyze independently — no coordination, no groupthink.
Round 2: Each agent reads the others' work and responds. Hard stop after Round 2.

↓

Synthesis + Audit

Violet synthesizes the result. Every output is audited against a 6-point checklist before it reaches the human. No unreviewed AI output goes out.

↓

Learn

New corrections, decisions, and insights are written to shared memory. Valuable knowledge fragments are queued for human review and compiled into the local model's memory kernel.

Dual Runtime

Violet runs on both cloud and local inference. The same agent identity, memory, and protocols — different compute.

Cloud

Claude via Claude Code

Full agent identities, multi-agent deliberation, complex analysis, MCP tool integration. Frontier-model reasoning with structured orchestration on top.

Anthropic Claude

Local

Ollama via Purple CLI

Condensed identities (<2K tokens), MCP tool support, offline-capable sovereign inference. Learns from cloud sessions without retraining — through the memory kernel.

Purple-Directive: CLI

Infrastructure

Sovereign hardware. No cloud vendor lock-in for compute. Encrypted mesh network, redundant edge tunnels.

Compute Node Inference + Training

Local inference Fine-tuning Private search Tunnel endpoint

Linux server · GPU accelerated

Always-On Node 24/7 Services

Web services Tunnel endpoint Monitoring

Low-power · always-on

Network Appliance Edge + Redundancy

DNS filtering VPN gateway Self-healing

ARM SBC · minimal footprint

Cloudflare Edge

HA tunnels Zero Trust Pages SSL

Managed edge — $0/mo

All nodes connected via encrypted mesh VPN. Redundant tunnel endpoints across multiple independent nodes ensure high availability.

Recent Updates

What's actually been built, in order. Not a roadmap — a log.

2026-03-04 Infrastructure

Dedicated 24/7 always-on node

Repurposed low-power hardware as a dedicated always-on server. Containerized web services, tunnel endpoints, and monitoring — freeing the main compute node for inference workloads only.

2026-03-03 Infrastructure

Tunnel redundancy across multiple nodes

Cloudflare Tunnel endpoints now run on three independent nodes. Redundant HA connections mean the site stays up even if two nodes go down simultaneously.

2026-03-03 AI

Violet unified workspace

Merged Project Amethyst (business ops) and Project Sapphire (personal AI) into a single Violet workspace. One COO. All domains. Unified memory, shared agent council, single source of truth.

2026-02-25 AI

Dashboard v2: live service status + STATE.md

Private dashboard rebuilt with HTMX partials — service grid refreshes every 15s, STATE.md renders as collapsible sections, teaching fragment review queue now has a web UI.

2026-02-20 AI

Chat webui: thinking traces + MCP tools

New chat interface with real-time WebSocket streaming, collapsible thinking trace rendering, code copy buttons, MCP tools connected, conversation history persisted locally.

All posts and updates →

What's Next

Active work and planned improvements.

live

24/7 always-on node Dedicated always-on hardware running all web services

in progress

PD logo — bubbly P/D monogram Custom mark where the outer curve reads P, inner hook reads D

in progress

Public demo: try Violet Restricted local model letting visitors experience the agent personality

planned

Purple Directive CLI v2 Rewrite with native tool calling, faster plan/build cycle

planned

Automated knowledge pipeline Overnight pipeline fully integrated — zero manual processing

planned

Multi-domain memory consolidation Single unified memory kernel across all operational domains

Open Source

The framework, protocols, and tools are public. Build your own agent organization.

Framework

Purple Directive: Violet

The full COO agent framework. E.I.K. advisory council, domain agents, memory kernel, teaching protocol, audit system, and all governing protocols.

View on GitHub →

Runtime

Purple Directive: CLI

Local AI runtime. Python app connecting to Ollama with MCP tools, condensed agent identities, plan/build modes. Offline-capable sovereign inference.

View on GitHub →