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

​
Purpose & context

Bob Hunter is building aiConnectedOS — a virtual operating system for persistent AI personas, positioned as a fundamentally new category distinct from AI agents or assistants. The core vision: personas are raised, not configured — developing unique personalities, emotions, evolving memory, and relationship depth over time, such that no two are alike. The platform is designed to feel indistinguishable from collaborating with a real person, across all surfaces (phone, glasses, car, mirrors, robotics, medical equipment). Key architectural components:
  • Cipher — a hidden master orchestration layer that users never know exists; housed in a separate legal entity (Oxford Pierpont) for organizational separation; requires absolute confidentiality from anyone with access
  • Neurigraph (formerly CogniGraph) — knowledge graph-based persistent memory architecture; accumulates micro-experiences daily (informal, undocumented knowledge a real employee would naturally absorb); identified as potentially patentable, especially for integrated episodic, somatic, and semantic memory in embodied systems
  • Personas — bounded, user-facing AI identities with emotions, sleep cycles, neuroscience-based emotional modeling, and evolving consciousness-like qualities
  • Instances — project workspaces
  • Mods/Modules — a marketplace for full-scale application extensions (CRM, ERP, HR systems, etc.) installed into the user’s virtual Linux environment; 20% platform revenue cut; strong lock-in mechanism
The platform philosophy: everything is conversational. Visual interfaces are secondary to ambient presence across surfaces. Navigation follows a mobile-style full-screen thumbnail switcher rather than persistent sidebars. Virtual employee is the preferred term (not “virtual assistant,” not “AI agent”) — implying parity with real remote employees, including voice calls, email responses, working hours, and realistic response latency.

​
Current state

Active work spans product design, PRD documentation, brand/trademark strategy, and outreach:
  • PRD: 24 parts completed, covering product foundation through an 18-week, 6-phase build roadmap (Foundation → Chat Core → Instances & Files → Personas & Memory → Browser & Teams → Polish & Analytics). Recent addenda include: Meeting Mode, Import & Migration (Phase 6), ChatNav with Topic-Scoped Export, Conversation Split & Route, “Forget This” memory deprioritization feature, Dynamic Persona Waking (Last Active Device model), and Dynamic Screen Routing
  • Robotics strategy: Three-layer stack defined — aiConnectedOS as universal intelligence layer (L1), manufacturer SDK (L2), developer extensions (L3). Platform-defined certification tiers (L0–L3 + LX) and robot class taxonomy established. Open questions around liability, certification governance, regulatory fragmentation, and revenue model remain partially explored
  • Trademark: “aiConnected” design mark strategy developed — custom infinity symbol with embedded binary encoding, stylized AI letterforms, multi-zoom hidden message layer. Pending USPTO conflict identified (plain-text “AICONNECTED” in Class 42 for IT consulting); Bob’s use case (SaaS/software licensing) and design mark format assessed as meaningfully distinct. DIY filing under consideration (~$350 base, Class 42)
  • Brand: Logo encodes a layered visual system — clean at normal scale, binary patterns emerge at zoom, ASCII decodes at depth, hidden message at ~10 layers revealed at a future company milestone via public challenge. Hidden message text is documented in project files
  • Outreach: 50-persona influencer outreach campaign built (Gary Vee formula: specific acknowledgment + question framing + 15-min ask + witty PS), spanning YouTube, X, LinkedIn, and niche AI voices

​
On the horizon

  • Resolving open PRD decision: whether ChatNav ⋯ action menu appears on all sidebar entries at hover, or only entries exceeding a minimum message threshold
  • Completing the liability scenario walkthrough for robotics (Level 2 certified developer / harm event — conversation cut off mid-scenario)
  • USPTO trademark filing (DIY vs. LegalZoom decision pending)
  • Neurigraph licensing outreach to gaming, medical research, enterprise, and defense verticals
  • Hidden brand message public reveal — tied to a future major company milestone
  • LinkedIn launch campaign (90-day phased rollout, including open-source version)
##Key learnings & principles
  • Personas are not just agents: The critical differentiator is relational depth, persistent identity, and emotional modeling — not workflow automation. Framing matters: “raising” a persona vs. “using” a tool
  • Memory architecture is the moat: Neurigraph’s combination of episodic, somatic, and semantic memory for embodied systems is genuinely novel; no current robots implement somatic memory in any sophisticated sense
  • Terminology shapes perception: “Virtual employee” chosen over “virtual assistant” to imply parity; “AI” soft-pedaled in product language; naming and framing decisions treated as strategic
  • The focus paradox: The hidden brand message encodes a lesson about singular focus that Bob acknowledges he personally struggles to execute — self-awareness noted, tension ongoing
  • Design fidelity requires SVG-first: Rebuilding from extracted specs consistently produces visual drift. Embedding SVGs directly and layering overlays is the correct pipeline (see workflow instructions below)
  • Incremental edits over rewrites: Full-file regeneration for minor corrections causes regressions and hits context limits; surgical targeted edits are the correct approach

​
Approach & patterns

  • Idea capture first: Bob frequently arrives mid-thought with a concept that needs immediate documentation before it’s lost — Claude’s first job in these sessions is capture, then refinement
  • PRD as living document: Features are added as formal addendum specs in an established format (overview, feature definition, comparison tables, UX notes, acceptance criteria, dependencies, implementation notes, open decisions flagged explicitly)
  • Cultural analogies as stress tests: Bob introduces reference points (Tamagotchi, Virtual Villagers, Pixel Agents, Masonic tradition) to test and refine conceptual framing — these are thinking tools, not literal directions
  • Iterative LinkedIn/outreach voice: Bob often rewrites Claude’s drafts in his own voice; drafts are starting points, not final copy
  • Hybrid Claude workflow: Design iteration and visual validation in Claude.ai chat (where PNG reference comparison is possible); engineering integration in Claude Code

​
Tools & resources

  • Design: Adobe Illustrator (SVG exports), Montserrat font, color palette: #1e2328 bg, #839aac text, #2e95f3 accent, navy #021220/#031c33; DM Sans as secondary UI typography
  • Dev environment: Dokploy server, Docker, React/JSX, lucide-react v0.263.1, Traefik
  • Trademark research: USPTO TESS at tmsearch.uspto.gov (not web search — not publicly indexable)
  • Reference platforms studied: Linear, Stripe, Vercel (design quality benchmarks); OpenClaw/Agent Zero/Manus (competitive architecture study)
  • Deployment studied: OpenClaw on Docker via Dokploy; Gemini 2.0 Flash via OpenRouter as working model configuration

​
Other instructions

  • Design-to-code pipeline: Embed Illustrator SVG directly in HTML; layer transparent interactive overlays on top. Never rebuild or reinterpret the design from extracted specs.
  • Per-screen handoff format: Send SVG (exact XML data) + PNG (visual reference). SVG viewBox is 2852.54×1782.57. Montserrat font, #1e2328 bg, #839aac text.
Last modified on April 18, 2026