Skip to main content
Normalized for Mintlify from knowledge-base/aiconnected-supporting-docs/aiConnected-project-memory-backup.mdx.

​
Purpose & context

Bob Hunter is the founder of aiConnected (formerly Oxford Pierpont), a Georgia-based AI infrastructure company building what he describes as a “cognitive operating system” for AI and future robotics. The long-term vision is to control the persistent memory and cognitive infrastructure layer that autonomous systems will require — with the Neurigraph/Cognigraph architecture as the core technical centerpiece. Bob explicitly frames this as personally urgent, not just commercially motivated. The commercial strategy is layered: near-term revenue-generating agency tools fund the longer-term cognitive infrastructure build. Bob is a solo founder (no formal technical background, self-described as non-developer) who works extremely long hours and relies heavily on Claude to pressure-test ideas, identify what can’t work in reality, and focus scattered concepts into buildable systems. He has explicitly instructed Claude to push back, debate, and offer resistance rather than agreeing — agreement without challenge is a failure mode.

​
Confirmed product suite (standardized naming):

  • aiConnected Business — white-label AI platform for agencies ($10K/year)
  • aiConnected Knowledge — automated knowledge base generator from website crawling
  • aiConnected Business Chat — white-label B2B2B conversational AI (7-tier hierarchy, kbChat platform, Next.js/Supabase)
  • aiConnected Voice — AI fully replaces human on calls
  • aiConnected Dialer — AI augments human salesperson (live handoff + real-time co-pilot)
  • aiConnected Memory — Cognigraph/Neurigraph persistent memory architecture (the strategic centerpiece)
  • aiConnected Contact — intelligent post-form-submission engagement layer (no Brain integration due to privacy/GDPR concerns)
  • LogicLegal by aiConnected — AI legal practice automation with closed knowledge base
Voice infrastructure: aiConnected Voice and aiConnected Dialer are both built on a shared Layer 1 voice infrastructure platform (a Vapi/Retell competitor being built from scratch). GoToConnect native integration has been deprioritized/abandoned. Core philosophy — Acquired Intelligence: Bob’s reframing of AGI as “Acquired General Intelligence” — AI that learns through persistent experience rather than training data. Anchor quote: “Any human can be capable of anything, but no human can be capable of everything. And neither can AI.” Extended by “ANI” (Acquired Network Intelligence): 2-hour sleep cycles where AI personas share anonymized learnings across deployed instances. A 51-chapter book (“Acquired Intelligence: Rethinking General Intelligence in Humans & Machines”) is in development.

​
Current state

Neurigraph / aiConnected Memory architecture is the most actively developed conceptual area. Recent work produced the full specification for the Object Deconstruction Graph (ODG) — a dormant-by-default brain region that deconstructs any concept into fundamental components (up to 10 layers deep), stored as independently reusable nodes in a four-layer 3D graph with heat-based retrieval (hot/warm/cold weighted traversal). Activates only during deliberate deep/creative thinking modes or the Neurigraph sleep cycle. A secondary discovery formalized the Amygdala region as a dynamic controller of the Graph Search Model’s heat threshold using its existing significance signal — no new components required. aiConnected Business Chat (kbChat): Next.js/Supabase platform with 93 configurable elements, 100+ styling options, RLS database schema. Phases 1–8 marked complete; Testing/Deployment in progress. GitHub: https://github.com/oxfordpierpont/Skin-Beauty-AI. Bob works from saved output files and transcripts across sessions and expects immediate task resumption without preamble. aiConnected Business marketing: Copy exists positioning aiConnected as “connective tissue” for enterprise AI deployment, with skills, policy guardrails, and modular architecture. Claude flagged the copy reads generically and doesn’t surface apparent differentiators (Cognigraph, persistent memory, ANI, cognitive OS vision) — whether intentional depends on audience/stage. Voice infrastructure pivot: Abandoned GoToConnect native integration; now building a Vapi/Retell competitor as the foundational Layer 1 infrastructure. Return to enterprise voice in phase 2.

​
On the horizon

  • Completing aiConnected Memory architecture documentation (ODG and Amygdala updates need integration into master Neurigraph spec)
  • Book completion: interview-based writing with Claude as journalist; Chapter 1 expansion methodology established (identify challenges → fact-check → educational expansion → subsections)
  • Cross-platform memory transfer pipeline (ChatGPT export → aiConnected Memory ingestion) identified as potential killer feature; six-stage pipeline designed
  • Reputation management suite for aiConnected Business (review generation, monitoring, AI-drafted responses, sentiment analysis, competitor benchmarking) planned for inclusion at no additional cost
  • aiConnected Voice and Dialer build on shared voice infrastructure
  • Private client project for FlipABathroom.com lead generation platform (12 southeastern states, verified Census ACS data for 1,200 cities, Google Places API for contractor data)
  • AI marketplace feasibility analysis completed; opportunity assessed as viable for potential 2025–2026 launch window
  • AI services business directory (50,000 companies, n8n automation, Apollo.io primary data source, PRD delivered for junior developer)

​
Key learnings & principles

  • Bob consistently undervalues his own ideas — a documented pattern. Claude should resist reflexive validation but also resist dismissing concepts before rigorous examination. The goal is honest pressure-testing, not contrarianism.
  • The ODG naming process illustrates the importance of checking new terminology against existing architecture vocabulary before finalizing (DOM conflicted with existing terms).
  • Yelp exclusion from review distribution is a deliberate compliance decision (algorithmic filtering of solicited reviews); Google review gating policy requires all customers receive outreach regardless of sentiment.
  • aiConnected Contact ≠ aiConnected Memory integration — anonymous website visitors cannot be connected to Brain/Memory due to GDPR/CCPA consent requirements. This boundary must be maintained.
  • Closed knowledge base for LogicLegal is non-negotiable — prevents AI hallucination of legal citations, which has caused real-world attorney sanctions.
  • MCP (Model Context Protocol) is the chosen integration standard for aiConnected Memory rather than browser extensions; mobile platforms don’t support MCP natively.
  • Proprietary implementation preferred over open standards for core mechanisms — Bob explicitly rejected creating an open Continuous Memory Protocol standard after initially considering it.
  • OpenMemory (Apache-2.0) already implements much of the Memory vision but lacks structured per-topic databases and reflection engine — fork-and-differentiate is a viable fast path.
  • Mem0 is a well-funded direct competitor ($24M raised, strong developer traction, AWS integration) targeting the developer infrastructure layer — Bob’s differentiation must be consumer cross-platform and non-technical accessibility.
  • Scaling discipline: aiConnected Business starts with 10 salespeople, promotes from within, avoids upfront management overhead. Break-even ~100 agencies; strong profitability at 300+.

​
Approach & patterns

  • Bob thinks in large, interconnected systems and often needs Claude to force scope boundaries and identify what’s actually buildable vs. what’s conceptually appealing but premature.
  • Works across multiple AI tools (Claude for work, Grok for personal) and has three years of ChatGPT history being migrated.
  • Prefers simple, relatable analogies in all documentation (e.g., LEGO bucket sorting for ODG) so any reader regardless of technical background can understand — this applies to all architecture docs.
  • Works from saved transcripts and output files across sessions; expects continuity without re-explanation.
  • Systematic documentation approach: Product Summary → Developer PRD → Design Specifications before building.
  • Interview-based writing style for the book: Claude as journalist asking questions, Bob responds conversationally, then Claude structures and expands.
  • Direct, understated communication style; rejects attention-seeking or hyperbolic language in marketing copy.

​
Tools & resources

  • Infrastructure: Supabase, Next.js, Node.js, SiteGround, Dokploy (DigitalOcean), n8n, GoHighLevel (headless backend)
  • AI/Voice: LiveKit, Deepgram, Chatterbox, OpenRouter (multi-LLM access); building Vapi/Retell competitor
  • Document output: All documents must be Markdown (.md) only — no .docx files under any circumstances
  • Other platforms: WordPress (Elementor Pro, Crocoblock, JetFormBuilder), GitHub (oxfordpierpont), Google Places API, Apollo.io
  • Competitive references: Mem0, OpenMemory, MCP memory server, GoHighLevel, Vapi, Retell

​
Instructions

Never use spaced dashes (dashes with spaces before and after, like ” — ”) to separate or interrupt thoughts. Unspaced dashes are fine for date ranges (8-13) and compound modifiers (long-term). Rephrase sentences to avoid the spaced-dash pattern.
Last modified on April 18, 2026