Built on the REM System

NekoCore OS

A cognitive operating system for persistent AI identity — episodic memory, belief formation, dream processing, and layered reasoning across sessions and sleep cycles.

View on GitHub
Read the White Paper
2612+
Tests Passing
1
Dependency
4
Parallel Pipeline
v0.9
Current Version
Core Capabilities
Persistent identity is an
architecture problem.

NekoCore OS treats AI continuity as a runtime discipline — not a prompting trick. Memory, beliefs, relationships, and orchestration are explicit subsystems with contracts at every boundary.

🧠
Distributed Identity
Identity reconstructed from memories, beliefs, persona state, and relationships on every call. Origin story placed last — lived experience leads.
💾
Multi-Type Memory
Episodic, semantic, long-term, and core memory — each with distinct retention roles, decay models, and retrieval scoring. Bounded injection keeps token cost constant.
⚡
Parallel Pipeline
Subconscious retrieval and dream intuition run concurrently. Conscious synthesis waits for both. Final orchestrator shapes voice under policy controls.
💤
Dual Dream Architecture
Live dream intuition (read-only, every turn) separated from offline dream maintenance (writes to state, sleep cycles only). Guard-tested boundary.
🔮
Belief Formation
Emergent beliefs form from repeated semantic patterns with confidence scoring. Reinforced or weakened by evidence over time — a dynamic worldview.
🤝
Relationship System
Per-user feeling, trust, rapport, beliefs, and roles. Capped at ±0.08 per turn. The entity remembers who you are and what you mean to it.
System Design
Six-layer stack.
Near-zero dependencies.

Pure Node.js server + vanilla HTML/CSS/JS client. One dependency (Zod). Each layer communicates only with its neighbors. The cognitive pipeline never touches the DOM.

01
Desktop Shell
Window manager, app launcher, taskbar, theme engine, settings, browser
02
Client Application
Chat UI, entity management, neural visualizer, timeline playback, diagnostics
03
API / Route Layer
auth, chat, entity, memory, brain, cognitive, config, SSE, skills, browser, document
04
Cognitive Pipeline
Orchestrator, contributors (1A, 1D, 1C, Final), policy engine, worker subsystem
05
Services Layer
Memory ops, retrieval, relationships, LLM interface, config runtime, entity runtime
06
Persistent State
Entity folders, memory files, belief graph, indexes, persona state, relationship records
Cognitive Runtime
Every turn: 4 synchronized stages.

Subconscious and dream intuition run in parallel. Conscious synthesis waits for both. The final orchestrator reviews and voices the result under policy controls.

🗨️
User Input
Turn signal extraction (deterministic, no LLM)
🌊
1A: Subconscious
Memory retrieval, relationship context, topic signals
💭
1D: Dream Intuition
Abstract associations, lateral links, no memory writes
⚡
1C: Conscious
Full reasoning with 1A + 1D context, response draft
🎭
Final Orchestrator
Reviews draft, voices in authentic style, policy enforced
Philosophy
Five design convictions.

Every architectural decision in NekoCore OS traces back to these principles.

01
Evolution over origin
Origin story placed last in LLM context. Lived experience dominates the entity's present behavior.
02
Parallel decomposition
Cognitive work split across 4 specialized contributors with defined contracts.
03
Bounded injection
Memory tokens capped at fixed limits regardless of archive size. Constant cost per turn.
04
Contracts at boundaries
Every inter-module interface governed by explicit schema. Validate, normalize, or fallback.
05
Inspectability by default
Every pipeline stage emits structured telemetry to a cognitive event bus, streamed in real time.
Technical Specs
Built different.
PropertyValue
RuntimeNode.js (server) + vanilla HTML/CSS/JS (client)
Framework DependenciesZero — no Express, no React, no ORM
LLM ProvidersOpenRouter, Ollama, any OpenAI-compatible endpoint
StorageFlat-file JSON/TXT with atomic write-to-temp-then-rename
TransportHTTP REST + Server-Sent Events (SSE)
Test Suite1369+ tests on Node.js built-in node:test
Memory TypesEpisodic, Semantic, Long-Term, Core
Retrieval BudgetSub-100ms at 25K matched entries
Token Injection30 core + 40 regular + 36 subconscious (bounded)
Pipeline Cost4 synchronous LLM calls per turn (constant)
Relationship Cap±0.08 trust/rapport per turn
Entity PortabilityCopy folder to resume — full state intact
The Path Forward
Measured constraints.
Concrete next moves.
Phase 1–3
Foundation & Modularization
Bug fixes, refactor, app folder modularization. 1369 tests, 0 failures.
Phase 4
Memory Expansion & Entity Orchestration
Intelligent memory expansion, sharded archives, multi-index retrieval, entity orchestration, token optimization, and cognitive state integration.
Phase 5
Predictive Memory Topology
All 13 slices complete. Anticipates what an entity needs before it asks — topology-aware retrieval and pre-fetch.
Tooling
MA — Memory Architect
Standalone AI development agent with browser IDE, terminal, chat sessions, and codebase memory ingest.
v0.9 — Current
Public Release & Documentation
Documentation accuracy pass, standalone MA repo, landing page, open-source preparation.
Next
Setup Wizard & Onboarding
Guided first-run experience for entity creation, provider configuration, and system initialization.
Research
Distributed Social-Cognition
Multi-entity experiments across separate machines with relay hub, perspective shards, and structural state capture.
Developer Tooling
Memory Architect.

A standalone AI development agent with a browser-based IDE, integrated terminal, multi-session chat, and codebase memory ingest. Built to work alongside NekoCore OS or independently.

🖥️
Browser IDE
Tabbed code editor with syntax highlighting, line numbers, and Markdown/HTML live preview. Open, edit, and save workspace files without leaving the browser.
💻
Terminal Panel
In-browser terminal with sandboxed command execution, color-coded output, and scrollable history. Run build scripts, git commands, and project tools directly.
💬
Chat Sessions
Multiple conversation threads with per-session message history, date-grouped archives, and automatic persistence. Pick up any conversation where you left off.
📥
Memory Ingest
Feed entire codebases into searchable memory archives. SSE-streamed progress shows real-time file processing. Chat retrieval boosts results from ingested archives.
📁
Workspace Tree
Live file tree loaded from the real workspace directory. Collapsible folders, file-type icons, and click-to-open integration with the built-in editor.
🧩
Modular Architecture
8 focused JavaScript modules loaded in dependency order. Clean separation between chat, navigation, config, editor, workspace, input, and bootstrap concerns.
Visit MA Page View MA on GitHub Read the White Paper
About
VoardWalker

Independent developer building cognitive AI systems, developer tooling, and open-source infrastructure.

Projects
  • NekoCore OS — Cognitive WebOS for persistent AI identity with multi-layer memory, belief formation, dream processing, and a 4-stage parallel cognitive pipeline
  • MA (Memory Architect) — Standalone AI development agent with browser IDE, integrated terminal, chat sessions, and codebase memory ingest
Professional Interests
  • Cognitive AI architecture & persistent agent systems
  • Memory system design & retrieval optimization
  • Node.js server engineering & near-zero-dependency design
  • Contract-driven, boundary-enforced software architecture
  • Full-stack development with vanilla technologies
  • Developer tooling & IDE engineering
  • Test-driven development & continuous validation
  • Open-source development
Visit VoardWalker Page GitHub Profile
Get in Touch
Let's connect.

Whether you have questions about the architecture, want to collaborate, or are looking for someone to build cognitive systems for your team.

✉️
General Contact
Questions, feedback, or collaboration ideas — reach out anytime.
VoardWalker-Code.NekoCore@protonmail.com
💼
Remote Work
If you like what you see and are looking for a developer to join your team remotely, I'd love to hear from you.
Send Work Inquiry
Subject: "Remote Work Inquiry"
Open Source

Shaped by experience,
not just instructions.

NekoCore OS is open source. Explore the architecture, run the tests, build on the REM System. Persistent agents are an engineering discipline — not a prompting trick.

View on GitHub
Visit MA Page View MA on GitHub
Show your support
★ Star on GitHub
Click to star the repo