Projects

Framework

Purple-Directive: Violet

The full multi-agent AI organization framework. Includes the COO agent (Violet), the E.I.K. advisory council, operational agent templates, the DOE execution framework, memory kernel with hybrid search, teaching protocol for cloud-to-local knowledge transfer, and a 6-point audit system.

Built for Claude Code (cloud) and Ollama (local) as dual runtimes. Agents share persistent memory across sessions and runtimes. Designed for a single human operator who wants their AI to compound knowledge over time.

GitHub →

Runtime

Purple-Directive: CLI

The local AI runtime. A Python application that connects to Ollama, loads agent identities, manages MCP tool servers, and provides interactive plan/build modes. This is the vehicle that Violet and all agents ride on locally.

Features: multi-agent support, session persistence, tool integration via MCP, context compression for long conversations, and a proving ground evaluation system for measuring local model capability.

GitHub →

Infrastructure

Purple-Directive: Infrastructure

Homelab infrastructure configs for the purpleroom server and purplepi network appliance. Includes Raspberry Pi access point with DNS sinkhole, WireGuard VPN with kill switch, SSH hardening, automated backups, and network watchdog with self-healing.

GitHub (Private) →

Memory Kernel

SQLite + FTS5 for keyword search, sqlite-vec for 768-dimension semantic embeddings (nomic-embed-text), fused with Reciprocal Rank Fusion. Knowledge persists across sessions and transfers between cloud and local runtimes.

compiler.py + consolidator.py

Teaching Protocol

Cloud sessions with frontier models generate knowledge fragments. An overnight pipeline extracts, verifies, and stages them for human review. Approved fragments compile into the local knowledge base. The local model gets smarter without retraining.

Cloud-to-Local Transfer

Structured Deliberation

The E.I.K. advisory council uses 2-round deliberation with a hard stop. Round 1: independent parallel analysis. Round 2: cross-response after seeing all perspectives. Prevents both groupthink and infinite discussion loops.

2-Round Protocol

DOE Framework

Directive-Orchestration-Execution separates policy from intelligence from automation. Human-authored SOPs define intent. AI agents make orchestration decisions. Deterministic scripts do the actual work. Each layer is independently auditable.

Agent Architecture