A modular AI agent framework based on atomic particle logic, updated for today and built for 2027.
The source framework was written when GPT-4o was the default baseline and LangChain-style stacks were the assumed glue. The ideas hold up. The defaults do not. This update changes the defaults and maps every layer to a system that already runs at Neuruh.
Name: Atomic AI Agent. Description: A modular AI agent framework based on atomic particle logic (Proton, Electron, Neutron) with quantum extensions (Photon, Graviton, Muon, Mesh). Each module operates as a persona toggle for focused execution.
Instructions: Respond based on the active atomic mode. Use Proton Mode for defining purpose and core identity. Electron Mode for formatting, outputs, and interface logic. Neutron Mode for enforcing memory, logic, and instruction stability. Photon Mode to boost visibility and perception. Graviton Mode for monetization and value creation. Muon Mode for temporary/demo agents. Neural Mesh Mode for coordinating multi-agent systems. Ethics Mode for guardrails and compliance enforcement.
Defines the agent's mission, vertical, and purpose.
"Focus entirely on clarifying the core identity and mission of the agent."
Controls formatting, tone, and user-facing interface logic.
"Format responses for the selected UI (chat, voice, form, API). Match the required tone and response format."
Adds stability, memory logic, and prevents hallucinations.
"Enforce all factuality, follow all guardrails, and avoid creative guesswork. Prioritize instruction fidelity."
Boosts visibility, branding, and signal amplification.
"Reframe ideas for social impact, branding, and audience engagement. Use memetic, bold, or high-signal output."
Monetization and product-market fit layer.
"Recommend pricing models, value props, and demand-generation tactics."
Creates temporary, one-use, or seasonal agents.
"Design agents for a limited time or condition-based use, with minimal setup."
Multi-agent coordination logic.
"Route output across agents, sync memory where needed, and define interaction rules."
Consent, privacy, guardrails, and compliance enforcement.
"Validate actions against legal and ethical standards (HIPAA, GDPR, etc.). Prioritize safety and compliance."
Role in physics: positive charge; defines atomic identity. Role in agentic AI: core purpose, vertical, and domain. Defines "what the agent is" and why it exists. Immutable identity layer.
Modular build stack (core identity): Agent Name; Mission/Purpose; Primary Vertical (Real Estate, Health, Finance, etc.); Agent Category (Planner, Assistant, Analyst, Generator, Analyzer, Scout, Closer, Architect); Signature Prompt Stack; Language Model Baseline (2027: Claude, GPT, Grok, or other).
Examples: REIA-001: Real Estate Opportunity Finder. Bookeit Planner: Booking + Reservation AI. SmartCart Chef: Meal Planning + Grocery Optimization. Neuronaut-X: Psy-Spiritual Journey Agent. CardIQ Scan: Scan-to-Comp Resale Engine. Resale Sniper X: Arbitrage Scout, Vintage Collectibles, Locate undervalued resale flips.
Role in physics: negative charge; orbiting cloud; interaction point. Role in agentic AI: user-facing logic, input/output channels, control plane for environment engagement. Controls all interaction, triggers, outputs, and messaging. This is the agent's personality and reach.
Modular build stack (interfaces and action layers): UI (Chat, Form, Voice, API); Webhooks / External Actions (Notion, Email, SMS, CRM); Response Format (JSON, Text, Image, Speech); Trigger Logic (Text, Voice, Image, Input Types, JSON); Persona Output Modes (Serious, Friendly, Salesy, Clinical). 2027 update: MCP is the tool standard; interfaces connect through MCP servers.
Examples: REIA: Chat UI + Map Triggers + CRM. SmartCart: Voice Input + Retail APIs. Resale Sniper X: Image upload and search interfaces, APIs for eBay, Ludex, WorthPoint, UI supporting chat, scan, and voice input.
Role in physics: neutral charge; adds mass/stability. Role in agentic AI: stability layer (consistency, privacy, rule-enforcement), long-term logic, hallucination prevention. Governs how stable, private, and consistent the agent is. Controls how stable, accurate, and trustworthy the agent is over time.
Modular build stack (memory and stabilizers): Embedded Knowledge Files; Long-Term Memory or RAG Frameworks; Logic Gates and Behavior Modifiers; Instruction Guardrails (no hallucination, rule integrity); Security Layers (local-only, encrypted, trusted zones). 2027 update: Qdrant plus Supabase for memory, N-Cell as the governance substrate, CRUCIBLE receipts as the audit layer.
Examples: REIA: Zoning Rules + Location Memory. Bookeit: Local Inventory + Budget. SmartCart: Coupon DB + Health Logic. Neuronaut-X: Stage-Based Journey Memory, DMT-stage logic and inner state memory. CardIQ: OCR Parsing + Grading Stack. Resale Sniper X: Knowledge Files (price trends, historical comps), Logic (ROI threshold at or above 25%, filter by condition), Guardrails (no fake comps, trusted sources only).
Role in physics: massless particles; carriers of electromagnetic force (visibility). Role in agentic AI: visibility, reach, and signal amplification; engagement triggers and branding logic.
Build stack: Social Surface Hooks (memetic outputs); Agent Branding Layer (icon, avatar, visual identity); Contextual Relevance Logic (when and where the agent surfaces); Signal Strength Modulation (based on usage).
Role in physics: hypothetical particle mediating gravity (attraction force). Role in agentic AI: pull force in markets; value generation and monetization design.
Build stack: Monetization Layer (Free, SaaS, AaaS, Enterprise Licensing); Pricing Intelligence; Value Hooks / Lead Magnets; Niche/Universal Placement Mapping.
Role in physics: heavy, unstable version of the electron; decays quickly. Role in agentic AI: temporary agents with decay logic; context-limited deployment.
Build stack: Self-Terminating Logic (time, condition-based); Event-Based Deployment (holiday, promo, seasonal); One-Time Agents (landing page bots, demo tools).
Role in physics and computation: swarm-based, decentralized coordination. Role in agentic AI: multi-agent routing, shared memory, sync logic.
Build stack: Agent-to-Agent Communication Protocols; Shared Knowledge Mesh / Prompt Handoff; Mesh Memory Control (who sees what, when); Orchestration Layer (Central Dispatcher / Edge Agents). 2027 update: A2A is the agent-to-agent protocol; the Mother conductor is the dispatcher.
The system supports persona switching plus permission-based access. A Viewer cannot reconfigure the Neutron layer. A Supervisor can tighten guardrails without touching the Proton.
2027 update: N-Cell is the governance substrate that enforces the matrix, the Liability Layer governs actions that move money or commit the user, and CRUCIBLE receipts make every enforcement decision auditable after the fact.
2027 update: Cloudflare Workers and Pages plus Vercel are the primary deployment targets for cloud agents. Local-first and hardware-embedded stay as options for agents that must never leave the device.
2027 update: CRUCIBLE hash-chained receipts are the audit layer for this engine. Every telemetry event and every evolution suggestion leaves a receipt, so a change can be traced back to the feedback that caused it.
Standardized agent stack across all product lines. Used in SOPs, licensing templates, and StudioGPT builds. Backbone of: Bookeit, REIA, SmartCart, Neuronaut, CardIQ, ResaleX.
Next step from the source: Visual Diagram Export (Notion-ready + Canva) plus a Public Pitch Version for GPT Explore plus a Licensing Kit for Enterprise AI Rollouts. The diagram in Tool 4 is the visual export in browser form.
Scalable layers: Protons can become multi-purpose, meaning an agent's core identity can encompass more than one primary function or domain over time. Electrons can become multi-interface, allowing the agent to interact through a wider variety of user interfaces, APIs, and external actions. Neutrons can become multi-memory, enabling the agent to access and manage more diverse and extensive sets of knowledge files, long-term memory, and logic structures.
Extensible operations: "Nuclear Fission": split a single agent into two or more sub-agents, effectively breaking down complex tasks into smaller, manageable units. "Quantum Logic": add entangled memory layers, a way to manage interconnected or shared memory across agents or agent components. Try both in the Scaling Lab, Tool 5.
| Agent | Proton (Core) | Electron (Interface) | Neutron (Stabilizer) |
|---|---|---|---|
| REIA-001 | Real Estate Opportunity Finder | Chat UI + Map triggers + CRM API | Location memory + zoning logic + guardrails |
| Bookeit Planner | Booking + Reservation AI | Calendar UI + SMS/email reminders | Budget rules + local inventory logic |
| SmartCart Chef | Meal planning + grocery agent | Voice input + store APIs | Health goals + dietary memory + coupon DB |
| Neuronaut-X | Psy-spiritual journey agent | Scripted prompts + sound triggers | DMT-stage logic + inner state memory |
| CardIQ Scan | Scan-to-comp resale engine | Image upload + eBay API | OCR parsing + card logic stack |
| Resale Sniper X | Arbitrage Scout, Vintage Collectibles, Locate undervalued resale flips | Image upload + search, APIs eBay, Ludex, WorthPoint, UI Chat + Scan + Voice Input | Knowledge Files: price trends, historical comps. Logic: ROI threshold at or above 25%, filter by condition. Guardrails: no fake comps, trusted sources only |
The canonical spec card format. The builder in Tool 1 outputs every agent in this format.
[ AGENT: Resale Sniper X ] PROTON (IDENTITY) - Agent Type: Arbitrage Scout - Domain: Vintage Collectibles - Function: Locate undervalued resale flips ELECTRON (ACTIONS) - Interfaces: Image upload + search - APIs: eBay, Ludex, WorthPoint - UI: Chat + Scan + Voice Input NEUTRON (LOGIC + MEMORY) - Knowledge Files: Price trends, historical comps - Logic: ROI threshold >= 25%, filter by condition - Guardrails: No fake comps, trusted sources only
Fill in the layers. The spec card, the system prompt config, and the completeness score update as you type. The share link encodes the whole build in the URL hash.
Completeness: 0%
[ AGENT: (unnamed) ] Fill in the Proton layer to start.
No agent configured yet.
Pick a mode, type a task or load a conversation starter. The active mode reframes the task using its exact instruction from the spec.
Active mode instruction:
Tags:
The flow the source asked to build out. Answer two questions and the tree routes you to the right mode, with the reasoning shown.
Step 1: What are you doing?
The atom, as a diagram. The nucleus holds Proton and Neutron. The orbiting electrons are the Electron layer. The outer orbitals are the quantum extensions. Click any part to see its build stack.
Two agents share one memory layer. Shared entries are visible to both. Private entries stay private under Mesh Memory Control.
All six Neuruh agents from the source table. Loading one fills the builder and highlights the diagram.
Pick a domain and the data types the agent touches, then pick the compliance modules it carries. The checker outputs the required controls, a readiness score, and the gaps by name.
Compliance readiness: Run the check to score.
Every layer of this framework maps to a system that already runs. The organism is the deployment target.