Research on grainworks

Memory Architecture Research

Wed, 10 Jun 2026 16:30:00 -0400

A deep-dive comparison of two SQLite-backed agentic memory systems for Hermes Agent: Sibyl Memory (the #2-ranked system on LongMemEval) and Holographic (our local-first baseline).

The Question

Agents need memory to work across sessions. But every memory system makes a tradeoff: cloud vs local, free vs paid, simple vs feature-rich. Which one is right for an offline-first, zero-tracking research setup?

Key Findings

Dimension	Holographic	Sibyl Memory
Storage	SQLite + FTS5 (3 tables)	SQLite + FTS5 (5 tiers)
Locality	Fully offline	Requires cloud activation
Cost	Free, unlimited	Free cap: 2 MB
Feature	Trust scoring, HRR vectors	Self-learning, 5-tier schema
Search	FTS5 + trust-weighted	FTS5 + multi-tier + LLM re-ranking
Dependencies	numpy (optional)	pip + cloud account

The Verdict

Sibyl wins on features — self-learning, multi-tier schema, validated retrieval. But the 2 MB free cap and cloud activation requirement are dealbreakers for an offline-first system.
Holographic wins on operational simplicity — zero dependencies, unlimited storage, no accounts. Less flashy but operationally bulletproof.

Where We Landed

We’re staying on Holographic for the orchestrator, research pipelines, and autonomous morning briefing. The self-learning gap is real, but not worth trading local-first operation.

Blumat Automatic Watering — Research

Wed, 10 Jun 2026 16:00:00 -0400

Status: Research compiled.

A technical deep-dive into Blumat automatic watering systems — the pressure-regulated drip irrigation systems that use ceramic soil moisture sensors instead of electronic timers. No power, no pumps, no controllers — just physics.

Key Areas

System architecture: How the ceramic cone sensors regulate water flow based on soil tension
Setup considerations: Pressure regulation, line runs, emitter placement, system bleeding
Integration potential: Adding IoT monitoring (moisture sensors, flow meters) to the purely mechanical system
Scale considerations: Small garden vs greenhouse vs market garden configurations

Source Files

Research doc: /Volumes/Mini_1Tb/Projects/research/blumat-watering-systems.md

Josh.ai — ELV Dealer Research

Wed, 10 Jun 2026 16:00:00 -0400

Status: Research complete. Briefing compiled.

A comprehensive deep-dive into the Josh.ai voice control platform — architecture, dealer onboarding, competitive positioning, and integration capabilities for the ELV (Electronic Lifestyles) market.

Research Scope

Platform architecture: Josh Core, Edge, Nano, Micro — what each processor does and how they differ
Integration ecosystem: Lutron, Sonos, AV systems, lighting, shading, HVAC, pools, security, and 50+ other integrations
Dealer training: Onboarding process, certification requirements, programming tools
Competitive analysis: Comparison with Control4, Crestron, Savant, and DIY alternatives
Market positioning: Where Josh.ai fits in the ELV landscape

Key Findings

Josh.ai’s strength is ease of programming and natural language voice control
The nano processor is the entry point; most dealers start with Core for whole-home
Lutron integration is the deepest and most mature
The platform is rapidly expanding its integration library (50+ partners as of 2026)

Source Files

Comprehensive briefing: /Volumes/Mini_1Tb/Projects/research/Josh.ai/JOSH_AI_COMPREHENSIVE_BRIEFING.md
167 source documents: PDFs, spec sheets, training materials, integration guides from the Josh.ai knowledge base
MEDITATION.md: Personal reflections on the research process
References: Individual integration docs in /Volumes/Mini_1Tb/Projects/research/Josh.ai/

Post-WWII Tech Boom — Economic Research

Wed, 10 Jun 2026 16:00:00 -0400

Status: Research compiled.

An analysis of the post-WWII technology boom: how massive wartime R&D investment in aerospace, electronics, materials science, and computing shaped the consumer technology landscape of the late 20th century.

Key Areas

The military-industrial-academic pipeline: How defense spending seeded commercial innovation (transistors, integrated circuits, jet engines, GPS, the internet)
The economic multiplier: How one dollar of wartime R&D produced tens of dollars of peacetime economic value
The seven patterns: Identified repeating patterns in how wartime technologies transition to civilian use
Parallels to AI: What the post-WWII boom tells us about the current AI investment cycle

Source Files

Research doc: /Volumes/Mini_1Tb/Projects/research/economic-cycles-knowledge-base.md
Seven patterns post: /Volumes/Mini_1Tb/Projects/research/seven-patterns-post.md
Three patterns deep dive: /Volumes/Mini_1Tb/Projects/research/three-patterns-deep-dive.md

SOUL.md — Trait Analysis & Personality System

Wed, 10 Jun 2026 16:00:00 -0400

Status: Active development. 8+ reference souls, trait-8000 analysis engine.

SOUL.md is an experimental framework for composing AI agent personalities. Instead of a single static persona, it defines agent souls as blends of trait vectors — letting them shift register, adopt archetypes, and compose new personalities from a library of reference souls.

The Framework

SOUL.md/
├── souls/ # Reference archetypes (persona SKILL.md files)
│ ├── jarvis/ # Calm, loyal guardian
│ ├── elizabethan-gentleman/ # Formal, witty
│ ├── eren-yeager/ # Intense, driven
│ ├── gojo/ # Playful, powerful
│ ├── niccolo-paganini/ # Artistic, eccentric
│ ├── rene-descartes/ # Philosophical, methodical
│ ├── senior-programmer/ # Technical, direct
│ ├── soldier-boy/ # Disciplined, loyal
│ └── rapper/ # Rhythmic, expressive
├── trait-8000/ # Trait analysis engine
├── assets/ # Visual references
└── souls/ # Full soul definitions

The Trait-8000 Engine

A Python-based analysis system that:

Sibyl vs Holographic — Memory at the Grain

Mon, 08 Jun 2026 16:00:00 -0400

Status: Published. Decision: staying on Holographic for now.

A head-to-head comparison of two SQLite-backed agentic memory systems for Hermes Agent:

Holographic: Local-first, zero-dependency memory provider with Phase Holographic Reduced Representations (HRR) for compositional retrieval. Unlimited database size, no accounts, no cloud roundtrips.
Sibyl Memory: Five-tier hierarchical schema (HOT/WARM/COLD/REFERENCE/ARCHIVE) with self-learning, LongMemEval-validated (95.6%), but requires cloud activation and has a 2 MB free cap.

The Verdict

Sibyl wins on features and benchmarks. Holographic wins on operational simplicity — zero external dependencies, unlimited storage, trust-based fact management.