NEXUS GEN3.AI  |  Complete Development Guide  |  CONFIDENTIAL










NEXUS GEN3.AI
COMPLETE DEVELOPMENT GUIDE
Enterprise AI Marketing Operations Platform




24 AI Agents  |  101+ Skills  |  14 Departments  |  9 Engines
23+ Chains  |  6 Heartbeats  |  3 Memory Tiers  |  Multi-Tenant SaaS


Founder: Med Talhi  |  gen3.ai  |  March 2026
Built on: OpenClaw + CrewAI + Postiz + AgentKits
CONFIDENTIAL - FOR DEVELOPMENT TEAM ONLY


Version 1.0 - Complete Reference Document
________________


Table of Contents


PART 1: EXECUTIVE SUMMARY & VISION
Chapter 1: Platform Overview
Chapter 2: Business Model & Pricing
PART 2: SYSTEM ARCHITECTURE
Chapter 3: 10-Layer Architecture
Chapter 4: Multi-Tenant Isolation
Chapter 5: Memory Architecture (3 Tiers)
Chapter 6: Task Queue & Priority System
PART 3: AI AGENT ARMY
Chapter 7: Orchestrator (9 Engines)
Chapter 8: CEO Agent (Strategic Director)
Chapter 9: 6 Director Agents
Chapter 10: 14 Worker Agents
Chapter 11: 3 Security Agents
Chapter 12: 101+ Skills Complete Catalog
PART 4: DATA FLOWS & WORKFLOWS
Chapter 13: Command Flow Hierarchy
Chapter 14: Cross-Agent Data Flows
Chapter 15: 23+ Task Chains Detailed
Chapter 16: Learning Loop (Larry Loop)
Chapter 17: Proactive Engine
PART 5: FRONTEND - ALL PAGES
Chapter 18: Public Pages (Landing, Register, Login)
Chapter 19: Client Dashboard - All 16 Pages
Chapter 20: Admin Mission Control - 6 Tabs
Chapter 21: Global UI Components
PART 6: BACKEND & INFRASTRUCTURE
Chapter 22: Backend Architecture & APIs (60+ Endpoints)
Chapter 23: Database Schema (25 Tables)
Chapter 24: WebSocket & Real-time System
Chapter 25: Security Architecture
Chapter 26: Deployment & DevOps
PART 7: INTEGRATIONS & EXPANSION
Chapter 27: 14 External Platform Integrations
Chapter 28: No-Code Skill Expansion
Chapter 29: Scaling Roadmap (10 to 10,000 Clients)
Chapter 30: Cost Optimization (7 Strategies)
PART 8: DEVELOPMENT PHASES
Chapter 31: Phase-by-Phase Build Plan
Chapter 32: Testing & QA Checklist
________________


PART 1
EXECUTIVE SUMMARY & VISION


________________


Chapter 1: Platform Overview
NEXUS GEN3.AI is a next-generation multi-tenant AI SaaS platform that allows businesses to deploy autonomous AI agents for marketing automation, customer support, content creation, sales operations, data analysis, and workflow automation.


1.1 What NEXUS Does
NEXUS replaces an entire marketing department with 24 AI agents working 24/7. Each agent specializes in a specific domain and has its own set of skills, AI model, and cost profile. Together they create content, manage ads, reply to comments, find leads, send emails, analyze data, and continuously improve through a weekly learning loop.


1.2 Key Metrics
Dimension
	Current State
	6-Month Target
	12-Month Target
	Users
	Early MVP
	1,000 paying
	10,000 paying
	AI Agents
	16 core + 8 new
	50 agents
	200+ agents
	Skills
	101+ skills
	300 skills
	1,000+ skills
	Uptime
	Development
	99.5%
	99.9%
	MRR
	Pre-revenue
	$30K
	$300K
	Infrastructure
	Single VPS
	Multi-region
	Global edge
	

1.3 Tech Stack Summary
Layer
	Technology
	Purpose
	Frontend
	Next.js 14 + TailwindCSS + React
	SSR/SSG Dashboard + Landing
	API Gateway
	Kong / Nginx + Custom OAuth
	Rate limiting, auth, routing
	Backend
	Node.js + Express + TypeScript
	Business logic, API endpoints
	AI Core
	OpenClaw + CrewAI + LangChain
	Agent lifecycle, task execution
	Queue
	BullMQ + Redis
	Task queue with 4 priorities
	Database
	PostgreSQL (25 tables)
	Operational DB with Row-Level Security
	Cache
	Redis
	Hot cache, sessions, rate limits
	Vector DB
	Qdrant / Pinecone
	Learning patterns, semantic search
	AI Models
	Claude + Gemini + Qwen (Ollama)
	Multi-model routing for cost optimization
	Deployment
	Docker + PM2 + Caddy
	Current: Single VPS (Ubuntu 24.04)
	________________


Chapter 2: Business Model & Pricing
2.1 Pricing Plans
Plan
	Price/mo
	Daily Tasks
	Daily Cost Limit
	AI Models
	Agents
	Starter
	$99
	50
	$2
	Gemini + Qwen (FREE)
	12
	Growth
	$199
	150
	$8
	+ Haiku
	18
	Pro
	$349
	500
	$20
	+ Sonnet
	22
	Enterprise
	$499
	Unlimited
	$50
	All models
	24
	

2.2 Cost Per Client (AI Expenses)
Average daily AI cost per client: ~$9.79/day = ~$300/month
With cost optimization strategies: reduced to ~$3-5/day = ~$100-150/month
Target gross margin: 50-75% depending on plan tier.


2.3 Revenue Projections
Clients
	MRR
	AI Cost/mo
	Infra Cost/mo
	Gross Margin
	10
	$2,490
	$1,500
	$50
	37%
	100
	$24,900
	$10,000
	$200
	59%
	1,000
	$249,000
	$60,000
	$800
	75%
	10,000
	$2,490,000
	$400,000
	$3,000
	83%
	

2.4 Seven Cost Optimization Strategies
1. Model Routing: 70% savings - cheap model for simple tasks, expensive for complex
2. Prompt Caching: 90% savings - Brand DNA cached by Anthropic
3. Batch Processing: 80% savings - 30 posts = 1 API call instead of 30
4. Semantic Cache: 50% savings - same question = cached free answer from Redis
5. Ollama Local: 100% savings - Qwen 2.5 runs free on server for simple tasks
6. Plan Limits: 85% savings - Starter plan uses only free models
7. Smart Cleanup: 20% savings - TTL on Redis, archive old DB data
________________


PART 2
SYSTEM ARCHITECTURE


________________


Chapter 3: 10-Layer Architecture
NEXUS is organized into 10 horizontal layers, from the external world down to monitoring. Each layer has a single responsibility and communicates with adjacent layers through REST APIs, WebSocket, or message queues.


Layer
	Name
	Components
	Connection
	0
	External World
	9 platforms (Instagram, TikTok, YouTube, Facebook, LinkedIn, Twitter, Google Ads, Meta Ads, WhatsApp)
	Webhooks + OAuth + REST APIs
	1
	Entry Points
	Frontend (Next.js), Mobile, Telegram Bot, Webhooks
	REST API + JWT
	2
	API Gateway
	JWT Validation, Rate Limiter, Tenant Router, Request Logger, CORS
	Validated Request
	3
	Orchestrator (Brain)
	9 engines: Task Router, Chain Manager, Context Builder, Model Router, Cost Controller, Quality Gate, Cron Scheduler, Proactive Engine, Learning Engine
	Reads/Writes Memory
	4
	Triple Memory
	Redis (Hot <1ms), PostgreSQL (Warm <50ms), Vector DB (Cold <100ms)
	Bidirectional
	5
	Task Queue
	BullMQ with 4 priorities: CRITICAL, HIGH, MEDIUM, LOW
	Tasks distributed
	6
	Agent Army
	24 agents at 4 levels: CEO (1), Directors (6), Workers (14), Security (3)
	Uses tools/skills
	7
	Skills Layer
	101+ skills organized by category (content, image, video, audio, analytics, CRM, ads, SEO)
	Results return
	8
	Output + Learning
	Content output + Quality scoring + Learning Loop storage
	Monitored
	9
	Heartbeats
	6 monitoring systems: Agent, Cost, Queue, Security, Integration, Memory
	24/7 monitoring
	

Side panels: Right side = 3 isolated companies (Multi-Tenant). Left side = 4 security layers.
________________


Chapter 4: Multi-Tenant Isolation
4.1 Isolation Strategy
NEXUS uses a hybrid isolation model that balances cost with security:
Isolation Tier
	Model
	Use Case
	Cost
	Shared Pool
	Row-level tenant_id in shared DB
	Starter plan (<10 agents)
	Lowest
	Schema Isolation
	Separate PostgreSQL schema per tenant
	Growth plan (10-50 agents)
	Medium
	Database Isolation
	Dedicated DB instance
	Enterprise plan (50+ agents)
	Highest
	VPC Isolation
	Dedicated cloud VPC
	Enterprise+ / White-label
	Custom
	

4.2 How 3 Companies Run in Parallel (Example)
One Orchestrator serves all tenants. tenant_id is injected into every query, every Redis key, every S3 path, and every Vector DB namespace.


Resource
	Acme Corp (Enterprise)
	TechStart (Pro)
	FoodZone (Starter)
	Agents
	24 (all)
	22
	12
	Models
	All (Sonnet+)
	Up to Sonnet
	Free only
	Daily Tasks
	Unlimited
	500
	50
	Daily Cost Limit
	$50
	$20
	$2
	Database
	acme_schema
	RLS rows
	RLS rows
	Redis
	acme:*
	tech:*
	food:*
	S3 Storage
	/acme/
	/tech/
	/food/
	Vector DB
	ns:acme
	ns:tech
	ns:food
	

CRITICAL: There is NO path between tenants. Even though the Orchestrator is shared, tenant_id separates everything. Cross-tenant access is detected by Guard Scanner and blocked instantly.
________________


Chapter 5: Memory Architecture (3 Tiers)
Tier
	Technology
	Speed
	What is Stored
	TTL/Cost
	Hot
	Redis
	<1ms
	Brand DNA cache, Semantic cache, counters, chain state, sessions, rate limits
	6h-12h TTL / FREE
	Warm
	PostgreSQL
	<50ms
	25 tables: Tenant, User, Agent, Content, Lead, Campaign, Task, Chain, Skill, Integration, AuditLog + Row-Level Security
	$20/mo
	Cold
	Qdrant/Pinecone
	<100ms
	Success patterns, timing patterns, audience preferences, content lessons, per-tenant + global anonymous
	$10/mo
	

5.1 How Agents Use Memory
Before every task, the Context Builder assembles a memory_context object:
8. Reads Brand DNA from Redis (Hot) - instant, always available
9. Queries PostgreSQL (Warm) for recent related content and history
10. Searches Vector DB (Cold) for learned patterns and success formulas
11. Combines everything into a rich context passed to the AI model
________________


Chapter 6: Task Queue & Priority System
6.1 BullMQ Queue Architecture
All tasks flow through a single BullMQ queue backed by Redis. Tasks are prioritized into 4 levels:
Priority
	When Used
	Examples
	SLA
	CRITICAL
	Instant processing
	Negative comment alert, crisis detection, webhook response
	<3 seconds
	HIGH
	User is waiting
	Create video, reply to DM, generate image
	<30 seconds
	MEDIUM
	Scheduled tasks
	Weekly plan generation, ad optimization, newsletter
	<5 minutes
	LOW
	Background work
	Analytics collection, reputation scan, cleanup
	<30 minutes
	

6.2 Queue Flow
12. User action or Cron trigger creates a task request
13. Orchestrator's Task Router identifies the correct chain
14. Cost Controller verifies budget availability
15. Task enters BullMQ with appropriate priority
16. Worker picks up the task and executes it
17. On completion: result stored, WebSocket notification sent, next chain step triggered
18. On failure: retry up to 3 times, then alert admin
________________


PART 3
AI AGENT ARMY


________________


Chapter 7: Orchestrator (9 Engines)
The Orchestrator is the code-based brain of NEXUS. It is NOT an AI agent - it is a Node.js service with 9 specialized engines. It costs $0 in AI fees because it runs entirely on server code.


Engine
	Function
	Speed
	Details
	Task Router
	Routes incoming tasks to correct chain
	<5ms
	Reads taskType, looks up Chain Registry, identifies the multi-step chain to execute
	Chain Manager
	Executes multi-step chains (23+)
	<10ms
	Manages chain state in Redis, handles step completion callbacks, passes results between steps
	Context Builder
	Builds context from 3 memory tiers
	<60ms
	Reads Brand DNA from Redis, queries PostgreSQL for history, searches Vector DB for patterns
	Model Router
	Picks cheapest adequate AI model
	<3ms
	Rules: Simple content=Haiku, Complex analysis=Sonnet, SEO/data=Gemini FREE, Privacy=Qwen local
	Cost Controller
	Enforces daily limits per plan
	<2ms
	Checks: task count vs plan limit, daily cost vs cost ceiling. Blocks if exceeded.
	Quality Gate
	Rejects score <60, retries up to 3x
	<5ms
	Scores content 0-100 using Gemini FREE. Below 60 = reject and retry with different prompt.
	Cron Scheduler
	11+ automated scheduled jobs
	exact
	Sunday 8AM: weekly plan. Friday 8PM: learning loop. Daily 6AM: analytics. Every 6h: ad optimization.
	Proactive Engine
	Anomaly + Opportunity + Crisis detection
	varies
	5 detectors: Anomaly (6h), Opportunity (12h), Timing (daily), Recycler (weekly), Crisis (instant)
	Learning Engine
	Larry Loop - weekly evolution
	Friday
	Extracts lessons from performance data, stores in Vector DB with confidence scores, feeds next week strategy
	________________


Chapter 8: CEO Agent (Strategic Director)
8.1 Agent Profile
Property
	Value
	Role
	CEO of the AI army. Sets weekly strategy. Distributes all tasks.
	AI Model
	Claude Sonnet 4
	Cost
	$0.05 per task
	Frequency
	Weekly (Sunday 8AM) + on-demand
	Reports to
	Orchestrator (code-based brain)
	Reads from
	Data Analyst (metrics) + Competitor Analyst (intel) + Brand DNA + Vector DB patterns
	Outputs
	week_plan.json with 5 priorities + 35 tasks distributed to all departments
	

8.2 Weekly Strategy Workflow (W1.1)
Trigger: Cron - Sunday 8:00 AM
19. Data Analyst collects all platform metrics (Gemini FREE)
20. Competitor Analyst scans competitor activity (Gemini FREE)
21. Strategic Director reads both reports + Brand DNA + Vector DB patterns
22. Strategic Director writes week_plan.json with: 5 strategic priorities, 35 content tasks distributed across platforms, budget allocation (Meta 40%, Google 30%, TikTok 20%, LinkedIn 10%), A/B test decisions
23. Orchestrator reads plan and distributes tasks to all departments
24. Client sees plan in Command Center under 'This Week Strategy'


8.3 Performance Review Workflow (W1.2)
Trigger: Cron - Friday 8:00 PM (Larry Loop)
25. Data Analyst compares this week performance vs last week
26. Strategic Director extracts lessons: 'Reels outperformed static 3x'
27. Learning Engine stores lesson in Vector DB with confidence score
28. Strategic Director adjusts next week strategy based on lessons
Result: Week 1 = 2% engagement -> Week 12 = 8%+ (system evolves automatically)


IMPORTANT: The CEO Agent does NOT give direct commands to other agents. It writes the plan (week_plan.json), and the Orchestrator reads the plan and distributes tasks. The CEO 'suggests', the Orchestrator 'executes'.
________________


Chapter 9: 6 Director Agents
Directors are department heads. They manage specific domains, use workers to complete tasks, and connect to external APIs.


9.1 Ads Manager (Advertising & Media)
Property
	Value
	Model
	Claude Sonnet 4
	Cost
	$0.04/task
	Frequency
	Every 6 hours (auto) + on-demand
	Skills (7)
	meta-ads-api, google-ads-api, tiktok-ads-api, linkedin-ads-api, budget-allocator, ab-test-engine, roas-calculator
	Uses
	Content Writer (ad copy), Creative Producer (ad visuals), Data Analyst (ROAS data)
	Connects to
	Meta Ads API, Google Ads API, TikTok Ads API, LinkedIn Ads API
	

Ad Optimization Workflow (every 6 hours)
29. Data Analyst pulls ROAS, CPC, conversions per campaign
30. Ads Manager analyzes: ROAS > 3x = scale budget +20%. ROAS 2-3x = maintain. ROAS < 2x = pause. ROAS < 1x = pause + alert human.
31. If decision = 'ab_test': Content Writer writes new ad copy (3 variants), Creative Producer generates new ad images (3 sizes), Ads Manager launches A/B test
32. Changes applied directly via platform APIs


9.2 CRM Manager (Sales & CRM)
Property
	Value
	Model
	Claude Haiku 4.5
	Cost
	$0.02/task
	Skills (6)
	whatsapp-api, email-sender, lead-scoring, pipeline-manager, churn-predictor, follow-up-scheduler
	Team
	Lead Hunter (Sonnet, $0.03/search, 8 skills: google-maps-scraper, linkedin-scraper, instagram-scraper, brightdata, lead-scorer, lead-enricher, industry-directories, company-profiler)
	

9.3 Email Manager (Email Marketing)
Property
	Value
	Model
	Claude Haiku
	Cost
	$0.018/task
	Skills (6)
	sendgrid-api, mailchimp-api, audience-segmentation, subject-ab-tester, drip-builder, newsletter-compiler
	Uses
	Content Writer (campaign copy), Data Analyst (open/click rates)
	

9.4 Influencer Manager
Property
	Value
	Model
	Haiku + TTS
	Cost
	$0.06/task
	Skills (6)
	persona-memory, voice-consistency, content-calendar, audience-analyzer, multi-persona, platform-optimizer
	Uses
	Content Writer (posts in persona voice), Creative Producer (persona-consistent visuals)
	

9.5 Community & Engagement (2 Agents)
Agent
	Model
	Cost
	Skills
	Community Manager
	Haiku 4.5
	$0.015/reply (134+/day = BUSIEST)
	sentiment-analyzer, reply-drafter, dm-responder, review-responder, lead-flagger, tone-matcher, crisis-escalator, auto-reply-modes
	Reputation Monitor
	Gemini Flash FREE
	$0.001/scan
	mention-scanner, review-tracker, sentiment-trend, alert-generator, google-reviews, social-listening
	

9.6 Client Success (3 NEW Agents)
Agent
	Model
	Cost
	Skills
	Onboarding Guide
	Haiku
	$0.03/new client
	guided-tour, account-connector, sample-content, feature-explainer, progress-tracker
	Success Manager
	Gemini FREE
	$0
	churn-predictor, health-scorer, upsell-recommender, milestone-tracker, satisfaction-survey
	Support AI
	Haiku
	$0.003/question
	faq-search, troubleshooter, ticket-creator, feature-explainer, escalation-router
	________________


Chapter 10: 14 Worker Agents
Workers are the execution army. They receive tasks from the Queue and use their skills to produce output.


Agent
	Model
	Cost/Task
	Skills Count
	Key Skills
	Serves
	Content Writer
	Haiku 4.5
	$0.003
	10
	social-captions, blog-articles, video-scripts, ad-copy-writer, email-copy, hashtag-generator, cta-writer, thread-writer, landing-page-copy, product-descriptions
	Ads, CRM, Email, Influencer, SEO
	Creative Producer
	Sonnet 4 + APIs
	$0.04-$3.00
	22
	dall-e-3, midjourney, kling-3.0, ffmpeg-video, d-id-avatar, google-tts, elevenlabs, whisper-stt, thumbnail-gen, carousel-builder (+ 12 more)
	Content, Ads, Influencer, Email, Web
	Brand DNA Analyst
	Sonnet 4
	$0.08
	7
	website-crawler, color-extractor, font-detector, voice-analyzer, audience-profiler, competitor-identifier, brand-dna-builder
	ALL 24 agents read Brand DNA
	SEO Expert
	Gemini FREE
	$0.002
	7
	keyword-tracker, serp-analyzer, technical-seo-audit, backlink-checker, schema-generator, meta-optimizer, content-seo-scorer
	Content, Web, Analytics
	Content QA
	Gemini FREE
	$0.001
	6
	grammar-checker, brand-compliance, tone-analyzer, plagiarism-detector, fact-checker, hashtag-validator
	Reviews ALL content output
	Data Analyst
	Gemini FREE
	$0.003
	10
	instagram-api, tiktok-api, youtube-api, facebook-api, google-analytics, kpi-calculator, trend-detector, anomaly-detector, report-generator, chart-builder
	CEO, Ads, Email, Influencer, Learning
	Competitor Analyst
	Gemini FREE
	$0.002
	6
	competitor-crawler, social-tracker, content-analyzer, strategy-extractor, comparison-reporter, opportunity-finder
	CEO, Proactive Engine
	Web Developer
	Sonnet
	$0.07
	8
	landing-page-builder, html-generator, css-optimizer, conversion-optimizer, schema-markup, page-speed-fixer, dns-manager, form-builder
	Landing pages, SEO
	Lead Hunter
	Sonnet
	$0.03
	8
	google-maps-scraper, linkedin-scraper, instagram-scraper, brightdata, lead-scorer, lead-enricher, industry-directories, company-profiler
	CRM Manager
	Localization Agent
	Haiku
	$0.005
	7
	darija-adapter, egyptian-adapter, gulf-adapter, french-adapter, cultural-calendar, currency-converter, locale-detector
	Content for MENA markets
	PR Agent
	Sonnet
	$0.05
	5
	press-release-writer, media-outreach, crisis-response, brand-reputation, public-statement
	Public relations
	________________


Chapter 11: 3 Security Agents
Security agents are HIDDEN from clients. They operate behind the scenes to protect the platform.


Agent
	Type
	Cost
	Frequency
	Function
	NEXUS Advisor
	Sonnet (AI)
	$0.10/30min
	Every 30 minutes
	Scans all 24 agents. Reports only to Admin. READ-ONLY - cannot modify anything. Produces 3-section report: positive discoveries, warnings, security alerts.
	MoltGuard
	Code-based
	$0
	On skill install
	Scans every new skill before installation. Checks: code safety, permissions, API access, data leaks. Blocks suspicious skills before they can run.
	Guard Scanner
	Code-based
	$0
	Every request
	Monitors ALL API calls in real-time. Detects: cross-tenant access attempts, DDoS patterns, suspicious API usage. Auto-blocks threats.
	________________


Chapter 12: 101+ Skills Complete Catalog
12.1 Skills by Category
Category
	Count
	Skills
	Content Writing
	10
	social-captions, blog-articles, video-scripts, ad-copy-writer, email-copy, hashtag-generator, cta-writer, thread-writer, landing-page-copy, product-descriptions
	Brand Analysis
	7
	website-crawler, color-extractor, font-detector, voice-analyzer, audience-profiler, competitor-identifier, brand-dna-builder
	SEO
	7
	keyword-tracker, serp-analyzer, technical-seo-audit, backlink-checker, schema-generator, meta-optimizer, content-seo-scorer
	Quality Control
	6
	grammar-checker, brand-compliance, tone-analyzer, plagiarism-detector, fact-checker, hashtag-validator
	Image Generation
	6
	dall-e-3, midjourney, nano-banana-pro, image-router, background-remover, image-upscaler
	Video Production
	7
	kling-3.0, ffmpeg-video, d-id-avatar, slideshow-maker, caption-engine, transition-effects, watermark-adder
	Audio
	4
	google-tts, elevenlabs, whisper-stt, music-library
	Creative Utility
	5
	thumbnail-gen, carousel-builder, story-creator, aspect-ratio-fitter, product-photoshoot
	Advertising
	7
	meta-ads-api, google-ads-api, tiktok-ads-api, linkedin-ads-api, budget-allocator, ab-test-engine, roas-calculator
	CRM
	6
	whatsapp-api, email-sender, lead-scoring, pipeline-manager, churn-predictor, follow-up-scheduler
	Lead Hunting
	8
	google-maps-scraper, linkedin-scraper, instagram-scraper, brightdata, lead-scorer, lead-enricher, industry-directories, company-profiler
	Community
	8
	sentiment-analyzer, reply-drafter, dm-responder, review-responder, lead-flagger, tone-matcher, crisis-escalator, auto-reply-modes
	Reputation
	6
	mention-scanner, review-tracker, sentiment-trend, alert-generator, google-reviews, social-listening
	Analytics
	10
	instagram-api, tiktok-api, youtube-api, facebook-api, google-analytics, kpi-calculator, trend-detector, anomaly-detector, report-generator, chart-builder
	Email
	6
	sendgrid-api, mailchimp-api, audience-segmentation, subject-ab-tester, drip-builder, newsletter-compiler
	Localization
	7
	darija-adapter, egyptian-adapter, gulf-adapter, french-adapter, cultural-calendar, currency-converter, locale-detector
	________________


PART 4
DATA FLOWS & WORKFLOWS


________________


Chapter 13: Command Flow Hierarchy
Understanding who controls whom is critical for the development team:


13.1 The Real Hierarchy
33. User (requests something via UI or API)
34. Orchestrator (decides which chain to execute - CODE, not AI)
35. Task Queue (BullMQ distributes tasks by priority)
36. Agents pick up tasks and execute them


KEY INSIGHT: The Orchestrator never gives commands directly to agents. It puts tasks in the Queue. Agents are workers that pick up tasks from the Queue. The CEO Agent (Strategic Director) has a special role: it writes the weekly plan, but the Orchestrator reads the plan and distributes the tasks. The CEO 'suggests', the Orchestrator 'executes'.


13.2 Who Requests Whom
Requester
	Requests From
	What Flows
	Strategic Director
	Data Analyst
	Performance reports (weekly metrics)
	Strategic Director
	Competitor Analyst
	Competitor reports (new campaigns, strategies)
	Ads Manager
	Content Writer
	Ad copy variants (3 headlines + description)
	Ads Manager
	Creative Producer
	Ad images (3 sizes: 1080x1080, 1200x628, 1080x1920)
	Ads Manager
	Data Analyst
	Campaign ROAS data
	CRM Manager
	Content Writer
	Follow-up messages (WhatsApp/email)
	CRM Manager
	Lead Hunter
	Discovered leads with scores
	Email Manager
	Content Writer
	Campaign copy + subject lines
	Influencer Manager
	Content Writer
	Posts in persona voice (not brand voice)
	Influencer Manager
	Creative Producer
	Persona-consistent visuals
	Content Writer
	Creative Producer
	Images/video for content that needs visuals
	Community Manager
	CRM Manager
	Flagged leads from comments (pricing inquiries)
	Brand DNA Analyst
	ALL 24 Agents
	Brand DNA JSON read before every task
	Data Analyst
	Learning Engine
	Performance data for weekly learning loop
	Reputation Monitor
	Proactive Engine
	Negative sentiment trends for crisis detection
	________________


Chapter 14: Cross-Agent Data Flows
14.1 Brand DNA Flow (Most Critical)
Brand DNA is the identity of each client. It is created by the Brand DNA Analyst on signup and refreshed monthly. EVERY agent reads it before EVERY task to stay consistent.


37. Brand DNA Analyst crawls client website, extracts colors, fonts, voice
38. Builds brand_dna.json with: brand colors (hex), fonts, voice tone (formal/friendly/humorous), target audience demographics, competitors, services
39. Stored in: Redis (Hot, 6h TTL) for fast access + PostgreSQL (permanent) for persistence
40. ALL 24 agents read Brand DNA from Redis before every single task


14.2 Content Production Flow
41. Strategic Director includes content tasks in week_plan.json
42. Content Writer receives tasks (20-50/day) and writes text
43. Creative Producer receives scripts and creates visuals
44. Content QA reviews: grammar, brand compliance, tone, plagiarism, facts
45. Quality Gate scores 0-100. Below 60 = reject and retry
46. Approved content appears in Studio Review Queue
47. Client approves/edits/rejects
48. Approved content scheduled via Post Bridge to 9 platforms


14.3 Lead Pipeline Flow
49. Community Manager detects pricing inquiry in comment: 'How much does this cost?'
50. Community Manager flags as lead (intent = pricing_inquiry)
51. Lead Hunter evaluates: match score = 82/100
52. CRM Manager creates lead in Pipeline -> 'New' column
53. CRM Manager schedules WhatsApp follow-up in 24 hours
54. Content Writer writes personalized follow-up message
55. If no response after 3 days: auto follow-up via email
________________


Chapter 15: 23+ Task Chains Detailed
15.1 Example Chain: TikTok Video Creation (7 steps)
Total cost: $0.223 | Time: 28 seconds | 4 agents involved | 0 human intervention until review


Step
	Agent
	Task
	Model
	Cost
	Time
	1
	Content Writer
	Write script + captions + hashtags
	Haiku
	$0.003
	3s
	2
	Creative Producer
	Generate 5 video frame images
	DALL-E 3
	$0.20
	8s
	3
	Creative Producer
	Generate voiceover
	Google TTS
	$0.00
	2s
	4
	Creative Producer
	Compile images + audio + music = MP4
	FFmpeg (local)
	$0.00
	5s
	5
	Creative Producer
	Add animated captions
	FFmpeg (local)
	$0.00
	3s
	6
	Virality Scorer
	Score virality 0-100
	Gemini FREE
	$0.00
	4s
	7
	Quality Gate
	Score 78 >= 60 = PASS
	Orchestrator
	$0.00
	<1s
	

15.2 Example Chain: Instagram Comment Auto-Reply (3 steps)
Total cost: $0.003 | Time: 3 seconds | 3 agents | 100% automatic
Step
	Agent
	Task
	1
	Community Manager (Gemini FREE)
	Analyze sentiment: positive + intent = pricing_inquiry
	2
	Community Manager (Haiku $0.003)
	Draft reply matching Brand DNA tone
	3 (conditional)
	Lead Hunter (Gemini FREE)
	Evaluate as lead, score 82/100, create in CRM Pipeline
	

15.3 Example Chain: Weekly Content Plan (6 steps)
Total cost: $0.75 | Time: ~5 minutes | 5 agents | 35 pieces of content ready
Trigger: Cron Sunday 8AM, runs in parallel for each active client (24 clients = 24 parallel chains)


15.4 All Chain Types
Chain Name
	Steps
	Agents
	Trigger
	create_social_post
	5
	Content Writer + Creative Producer + QA
	User click
	create_tiktok_video
	7
	Content Writer + Creative Producer + QA
	User click
	create_avatar_video
	6
	Content Writer + Creative Producer + D-ID
	User click
	weekly_content_plan
	6
	Data Analyst + Competitor + CEO + Content Writer + Creative + QA
	Cron Sunday 8AM
	auto_reply_comment
	3
	Community Manager + Lead Hunter
	Webhook (instant)
	lead_discovery
	4
	Lead Hunter + CRM Manager
	User click
	auto_follow_up
	3
	CRM Manager + Content Writer
	Condition (3 days silence)
	ad_optimization
	4
	Data Analyst + Ads Manager + Content Writer + Creative
	Cron every 6h
	weekly_newsletter
	3
	Data Analyst + Content Writer + Email Manager
	Cron Thursday 5PM
	welcome_drip
	4
	Content Writer + Email Manager
	Webhook (new subscriber)
	influencer_calendar
	4
	Data Analyst + Influencer + Content Writer + Creative
	Cron Sunday 9AM
	brand_dna_extract
	7
	Brand DNA Analyst
	New client signup
	reputation_scan
	3
	Reputation Monitor
	Cron hourly
	seo_audit
	5
	SEO Expert + Web Developer
	Cron weekly
	performance_review
	4
	Data Analyst + CEO + Learning Engine
	Cron Friday 8PM
	________________


Chapter 16: Learning Loop (Larry Loop)
16.1 How the System Gets Smarter
Week
	Patterns
	Strategy
	Engagement
	Week 1
	None yet
	Generic: 50% images, 50% video
	2%
	Week 4
	Reels outperform +87%
	70% Reels, 30% images. lesson: {reels_win, confidence: 0.82}
	4.5%
	Week 8
	25 patterns discovered
	Optimized timing, format, hashtags, CTAs
	6.2%
	Week 12+
	45+ patterns
	100% customized per audience. Self-evolving.
	8%+
	

16.2 Learning Loop Mechanism
56. Friday 8PM: Data Analyst collects week performance vs last week
57. CEO Agent extracts lessons: 'Reels outperformed static 3x', '7PM posting = best for this audience'
58. Learning Engine stores each lesson in Vector DB with confidence score (0.0 - 1.0)
59. Lessons with confidence > 0.7 are used in next week strategy
60. Low confidence lessons are tracked but not yet applied
61. Over time, confidence scores increase as patterns are confirmed


16.3 Vector DB Pattern Storage
Each pattern stored as: {lesson: 'reels_win', confidence: 0.82, tenant_id: 'acme', context: 'instagram_reels', evidence: '87% more engagement than static', week: 4}
Patterns are scoped: per-tenant patterns are private, anonymized high-confidence patterns are shared globally.
________________


Chapter 17: Proactive Engine
The Proactive Engine does not wait for user commands. It monitors and acts autonomously.
Detector
	Frequency
	What It Detects
	Action
	Anomaly Detector
	Every 6 hours
	Current engagement < 30-day avg by 20%+
	Alert + suggest solution + 'Apply' button
	Opportunity Hunter
	Every 12 hours
	Competitor launched successful campaign
	Suggest counter-campaign with estimated impact
	Optimal Timing
	Daily
	Best posting time changed
	Auto-adjust schedule
	Content Recycler
	Weekly
	Old content had excellent performance
	Suggest: 'Convert to 3 Reels + thread + newsletter?'
	Crisis Detector
	Instant (real-time)
	5+ negative comments in 1 hour
	Immediate alert + suggest: unified response / pause ads / apology post
	________________


PART 5
FRONTEND - ALL PAGES


________________


Chapter 18: Public Pages
18.1 Landing Page (/)
The first impression. Premium marketing site with live AI demo.
* Hero: Dark background + animated particles + 'Your AI Marketing Army' heading
* Live counter: '847,293 tasks completed today' with counting animation
* CTA: 'Start Your 14-Day Free Trial' -> /register
* Features showcase: 6 cards (Content, Influencer, Ads, Social, Analytics, CRM)
* How it Works: 4-step animated flow
* Pricing: 5 plan cards (Starter $99, Growth $199, Pro $349, Enterprise $499, White Label $999+)
* Testimonials carousel + FAQ accordion + Footer


18.2 Registration (/register) - 5-Step Wizard
Step
	Title
	Fields / Actions
	1
	Basic Information
	Company Name, Your Name, Email, Password (strength indicator), Phone (optional), Country dropdown
	2
	Brand Analysis (AI-Powered)
	URL input -> Brand DNA Agent activates -> animated 6-step progress -> Brand DNA card with editable colors/fonts/voice
	3
	Goals Selection
	5 sections as selectable cards: Goal, Volume, Platforms, Budget, Timeline
	4
	Digital Influencer Setup
	Name, Avatar upload/AI generate, Voice dropdown with play button, Personality tags, Content pillars, Bio
	5
	Review & Activate
	Summary cards, Plan selection, Payment section (Stripe/CMI), 'Activate' button
	

18.3 Login Pages
* /login - Client login: Email + password + 'Forgot Password' link
* /admin/login - Admin login: Separate form for platform administrators
* /forgot-password - Email input + password reset flow
________________


Chapter 19: Client Dashboard - All 16 Pages
19.1 Global UI Components (Present on ALL pages)
* Sidebar: 260px expanded / 72px collapsed. 16 navigation items. Dynamic items per plan (lock icon for unavailable). 'NEW' badge on recent features.
* Header: Language switcher (EN/AR with RTL flip), Notifications bell with count, User avatar dropdown, '+ New Task' button, 'Weekly Report' button
* Floating Chat Widget: NEXUS Assistant on every page. Voice input. Command execution.
* Onboarding Tour: 6-step spotlight tour on first visit
* Powered By Badge: Shows which AI agents power each page


19.2 Command Center (/dashboard)
The home dashboard. Shows everything at a glance.
* 6 KPI cards: Posts published, Total views, Active clients, Revenue, Active ads, Active agents
* AI Insights section: 4 insight cards from Data Analyst with 'Take Action' links
* This Week Strategy: 5 bullet points from Strategic Director (from week_plan.json)
* Today's Trends: 3 trend idea cards with Brand DNA match % and 'Execute Now' buttons
* Today's Schedule: Table (time, platform icon, task, status)
* Team Activity Feed: Real-time agent activity log (right panel)
* Platform Performance: Cards for Instagram, Facebook, TikTok, YouTube, LinkedIn


19.3 Content Studio (/dashboard/studio)
Where ALL content is created and reviewed.
* 2 tabs: 'Create' and 'Content Queue'
* Create tab: 5 video type cards (Screen Recording $4, Card Reveal $5, Split Screen $6, Avatar Video $8, Article to Video $6). Each opens 4-step creation modal.
* Image Generator: Description input, style selector (8 styles), size selector, Generate button
* Content Queue tab: 'Ready for Review' cards with Approve/Edit/Reject + 'Approve All' batch button
* Content Queue: 'Scheduled' calendar (weekly, color-coded by platform)
* Content Queue: 'Published' history with engagement stats per piece
* Content Queue: Settings panel (frequency, platforms, content mix, tone, auto-approve toggle)


19.4 Digital Influencer (/dashboard/influencer)
Manage AI-powered digital influencer persona. Pro+ only.
* 3 tabs: Profile, Production Studio, Performance
* Profile: Persona card (name, avatar, voice, personality tags, stats), Edit Persona modal
* Production Studio: 4 production cards (Avatar Video, Voice Post, Image Post, Story Series)
* Performance: Follower growth chart, engagement by content type, top content, demographics


19.5 Agents Team (/dashboard/agents)
View and manage all AI agents assigned to this account.
* Grid of agent cards (16 agents, 4 columns desktop). Each card: name (AR+EN), role, status dot, current task, category label, progress bar, stats
* Click card -> Agent detail page with 5 tabs: Overview, Tasks, Performance, Settings, Logs
* Agent Performance Comparison section at bottom


19.6 Advertising (/dashboard/ads)
* Campaign table: name, platform badge, budget, spent, ROAS (colored), status, actions
* 'Create Campaign' -> 5-step wizard (platform -> goal -> budget -> audience -> content -> review)
* A/B Testing toggle in creation wizard
* AI Recommendations panel: 3 cards with Apply/Dismiss
* Auto-Optimization toggle with status badge


19.7 Social Media (/dashboard/social)
* Post composer: AI Generate button (3 options), platform checkboxes, date/time picker with 'Best Time' AI suggestion
* Monthly content calendar: colored dots per platform
* Recycled Past Hits: Top 3 posts with 'Recycle' button
* Smart Hashtag Generator: 15 suggestions with relevance scores
* Unified Inbox tab: All comments/DMs/reviews from all platforms. AI suggested replies. Auto-Reply modes (Manual/Semi-Auto/Full Auto)


19.8 Email Marketing (/dashboard/email)
* 5 tabs: Campaigns, Templates, Contacts, Analytics, Settings
* Campaigns: Table + 'Create Campaign' wizard (5 steps)
* Templates: 8 pre-built + 'Generate New'
* Analytics: 4 KPIs + campaign comparison + best times heatmap


19.9 Customers & Sales CRM (/dashboard/crm)
* 3 tabs: Pipeline, Contacts, Lead Discovery
* Pipeline: Kanban 6 columns (New, Contacted, Qualified, Proposal, Negotiation, Won) with drag-and-drop
* Lead Discovery: AI-found leads with match scores, 'Add to CRM' buttons, search settings panel


19.10 Analytics (/dashboard/analytics)
* Time filters: 7d / 30d / 3mo / year
* 4 KPI cards + Reach & Engagement trend chart + Platform comparison table
* SEO tab: Score gauge, keyword rankings, site audit, AI recommendations
* Brand Radar tab: Sentiment gauge, word cloud, mentions timeline, crisis alerts
* Reviews tab: Average rating, reviews feed with AI Reply buttons


19.11 Additional Pages
Page
	Path
	Key Features
	Reports
	/dashboard/reports
	Weekly auto-generated report + Custom report builder + PDF export
	Knowledge Base
	/dashboard/knowledge
	Brand DNA card + Voice examples + Competitors + Template library
	Landing Pages
	/dashboard/landing-pages
	AI page builder + 6 templates + domain connection
	Integrations
	/dashboard/integrations
	14 platform cards with OAuth2 connect flow
	Smart Workflows
	/dashboard/automations
	6 automation templates (toggle ON/OFF) + Visual workflow builder (React Flow)
	Help Center
	/dashboard/help
	Quick start guide + Tutorials + FAQ + Contact support
	Settings/Profile
	/dashboard/settings/profile
	Avatar, name, email, password, language, timezone
	Settings/Team
	/dashboard/settings/team
	Team members table + invite (Growth+)
	Settings/Billing
	/dashboard/settings/billing
	Plan card + payment method + invoices + usage meters
	Settings/Security
	/dashboard/settings/security
	2FA + active sessions + login history
	Settings/Notifications
	/dashboard/settings/notifications
	Toggle matrix: notification types x channels
	Settings/API Keys
	/dashboard/settings/api-keys
	API key + webhook URL + usage stats (Enterprise+)
	________________


Chapter 20: Admin Mission Control - 6 Tabs
The most important page. One page with 6 tabs that controls EVERYTHING.


Tab 1: System Overview
* 6 status cards: Server health (CPU/RAM/Disk), Services status (6 green dots), Active agents (13/16), Revenue today, Tasks processed, Error rate
* Live Agent Activity: Real-time animated data flow showing tasks being processed
* NEXUS Advisor section: 3 latest recommendations with Approve/Reject buttons
* Audit Log: Immutable table with timestamp, event type, description, user/agent, IP


Tab 2: Agent Management
* Table: All 24 agents with: name, model, status, success rate, tasks/day, cost/day
* Click agent -> Panel: Model config, cost controls, skill assignments, client assignments, performance metrics, task queue, logs, emergency controls (pause/restart/rollback)


Tab 3: AI Model Router
* 4 consumption cards: Claude Sonnet (tokens/cost), Haiku (tokens/cost), Gemini Flash (free tier remaining), Qwen (local free)
* Routing Rules Table: task type -> model -> reason -> cost
* Savings Calculator: 'Cost without routing: $5,000/mo. With routing: $1,200/mo. Savings: $3,800 (76%)'
* Manual Override: Force all tasks to specific model (testing) + per-client overrides


Tab 4: Feature Control
* Table: Feature name, client page path, agents involved, APIs used, plan availability toggles (Starter/Growth/Pro/Enterprise)
* Click any feature row: Preview (iframe), Workflow, Agents, APIs, config


Tab 5: Client Management
* Client table: company, plan, MRR, agents active, tasks/day, AI cost, status
* Click client -> Full detail: Brand DNA, usage stats, agent activity, billing, impersonate button


Tab 6: Revenue & Analytics
* MRR/ARR charts, Revenue by plan, AI cost breakdown, Margin analysis, Growth projections
________________


PART 6
BACKEND & INFRASTRUCTURE


________________


Chapter 22: Backend Architecture & APIs (60+ Endpoints)
22.1 Project Structure
src/
  index.ts              - Express app entry point
  config/               - database.ts, redis.ts, env.ts
  middleware/            - auth.ts, rateLimit.ts, errorHandler.ts, validators.ts
  routes/               - 15 route files (auth, dashboard, studio, social, crm, analytics, ads, email, agents, skills, automations, settings, reports, knowledge, admin)
  services/             - Business logic (auth, agent, skill, studio, social, crm, analytics, email, ai)
  websocket/            - Socket.io setup + channels (tasks, notifications, agents, chat, studio)


22.2 API Endpoints by Domain
Domain
	Prefix
	Key Endpoints
	Auth
	/api/v1/auth
	POST /register, POST /login, POST /forgot-password, POST /reset-password, GET /me, POST /refresh-token
	Dashboard
	/api/v1/dashboard
	GET /stats, GET /schedule, GET /activity, GET /insights, GET /strategy, GET /trends
	Studio
	/api/v1/studio
	POST /create (video/image), GET /queue, PATCH /queue/:id (approve/reject), GET /published, GET /scheduled
	Social
	/api/v1/social
	POST /compose, GET /calendar, GET /inbox, POST /inbox/:id/reply, PATCH /auto-reply-mode, GET /hashtags
	CRM
	/api/v1/crm
	GET /pipeline, PATCH /pipeline/:id/stage, POST /contacts, GET /leads, POST /leads/search, POST /leads/:id/add
	Analytics
	/api/v1/analytics
	GET /overview, GET /platforms, GET /seo, GET /brand-radar, GET /reviews, GET /insights
	Ads
	/api/v1/ads
	GET /campaigns, POST /campaigns, PATCH /campaigns/:id, GET /campaigns/:id/stats, POST /campaigns/:id/ab-test
	Email
	/api/v1/email
	GET /campaigns, POST /campaigns, GET /templates, POST /templates/generate, GET /contacts, GET /analytics
	Agents
	/api/v1/agents
	GET /, GET /:id, PATCH /:id/settings, GET /:id/tasks, GET /:id/performance, GET /:id/logs
	Admin
	/api/v1/admin
	GET /system-status, GET /agents, PATCH /agents/:id, GET /clients, GET /revenue, GET /model-usage, GET /advisor, GET /audit-log
	________________


Chapter 23: Database Schema (25 Tables)
PostgreSQL with Row-Level Security (RLS). Every table includes tenant_id for multi-tenant isolation.


Table
	Key Columns
	Purpose
	tenants
	id, name, domain, plan, status, settings_json, created_at
	Company accounts
	users
	id, tenant_id, email, password_hash, role, name, avatar, language
	User accounts with tenant association
	agents
	id, tenant_id, name, model, status, config_json, skills[], cost_today
	AI agent instances per tenant
	skills
	id, name, source, status, security_status, config_json, usage_count
	Skill registry (global)
	agent_skills
	agent_id, skill_id, enabled
	Many-to-many agent-skill assignment
	tasks
	id, tenant_id, agent_id, chain_id, type, status, input, output, cost, duration
	Every task executed
	chains
	id, tenant_id, name, steps_json, status, current_step, started_at
	Multi-step chain execution state
	content
	id, tenant_id, type, platform, text, media_url, status, ai_score, engagement_json
	All produced content
	campaigns
	id, tenant_id, platform, name, budget, spent, roas, status, settings_json
	Ad campaigns
	leads
	id, tenant_id, name, email, phone, source, score, stage, notes
	CRM leads
	contacts
	id, tenant_id, name, email, company, phone, tags[], last_contact
	CRM contacts
	integrations
	id, tenant_id, platform, access_token, refresh_token, expires_at, status
	Platform connections
	brand_dna
	id, tenant_id, colors[], fonts[], voice, audience, competitors[], services[]
	Brand identity per tenant
	notifications
	id, tenant_id, user_id, type, title, body, action_url, read, created_at
	User notifications
	audit_log
	id, tenant_id, event_type, description, user_id, agent_id, ip, created_at
	Immutable audit trail
	automations
	id, tenant_id, name, trigger, steps_json, status, last_run, next_run
	Smart workflow definitions
	email_campaigns
	id, tenant_id, name, subject, body, segment, status, stats_json
	Email campaigns
	landing_pages
	id, tenant_id, name, slug, html, visits, conversions, status
	Generated landing pages
	reports
	id, tenant_id, type, period, data_json, created_at
	Generated reports
	subscriptions
	id, tenant_id, plan, stripe_id, status, current_period_start, current_period_end
	Billing subscriptions
	invoices
	id, tenant_id, amount, status, stripe_invoice_id, pdf_url, created_at
	Payment invoices
	________________


Chapter 24: WebSocket & Real-time System
24.1 Socket.io Channels
Channel
	Events
	Purpose
	tasks
	task.started, task.completed, task.failed
	Real-time task progress updates
	notifications
	notification.new
	Push notifications to user
	agents
	agent.status_changed
	Agent status changes (active/idle/error)
	chat
	assistant.message
	NEXUS Assistant chat responses
	studio
	production.progress
	Content production progress (video encoding, image generation)
	

24.2 Connection Flow
62. Frontend connects: wss://test.gen3.ai/ws with JWT token in handshake
63. Server verifies JWT, extracts tenant_id, joins tenant-specific room
64. All events scoped to tenant room (tenant_id isolation)
65. On task completion: update task status in DB, emit WebSocket event, create notification
________________


Chapter 25: Security Architecture
25.1 Four Security Layers
Layer
	What
	How
	API Gateway
	JWT + Rate Limiting
	JWT with 24h rotation. Rate limits per plan tier. Request logging.
	Tenant Isolation
	Row-Level Security
	tenant_id on every query. RLS policies in PostgreSQL. Redis key prefixing.
	Agent Security
	3 Guard Agents
	NEXUS Advisor (30min scan), MoltGuard (skill scanning), Guard Scanner (real-time API monitoring)
	Data Security
	Encryption + Backup
	TLS everywhere. Passwords bcrypt. API keys encrypted. Daily backups.
	

25.2 Six Heartbeat Monitors
Heartbeat
	Frequency
	Monitors
	On Failure
	Agent
	30 seconds
	Is each agent alive? Success rate?
	Auto-restart, then alert admin
	Cost
	5 minutes
	Any client overspending (+50%)?
	Pause tasks + alert
	Queue
	10 seconds
	Queue length? Stuck tasks > 5min?
	Auto-scale workers
	Security
	1 minute
	Intrusion? Cross-tenant? DDoS?
	Block IP + alert
	Integration
	15 minutes
	Platform APIs up? Tokens valid?
	Auto-renew token
	Memory
	5 minutes
	Redis > 80%? DB slow queries?
	Clean old keys
	________________


Chapter 26: Deployment & DevOps
26.1 Current Server
Property
	Value
	IP
	76.13.63.240
	OS
	Ubuntu 24.04.4 LTS
	Project path
	/opt/nexus
	Domain
	test.gen3.ai
	Reverse Proxy
	Caddy (auto HTTPS)
	Process Manager
	PM2
	Frontend port
	3000
	Backend port
	3001
	

26.2 Scaling Roadmap
Scale
	Infrastructure
	Workers
	Cost/mo
	10 clients
	1 VPS, 5 Workers, Redis 500MB, PostgreSQL 1 instance
	5
	~$50
	100 clients
	2 VPS, 20 Workers, Redis 2GB, PostgreSQL + 1 Read Replica
	20
	~$200
	1,000 clients
	Kubernetes 5 nodes, 100 Workers, Redis Cluster, PgBouncer
	100
	~$800
	10,000 clients
	Multi-region K8s, 500 Workers, Redis Cluster (3 nodes), PostgreSQL Citus
	500
	~$3,000
	________________


PART 7
INTEGRATIONS & EXPANSION


________________


Chapter 27: 14 External Platform Integrations
Platform
	Integration Type
	Features
	Auth
	Instagram Business
	API + Webhooks
	Publish, analytics, comment replies, DM responses
	OAuth2
	Facebook Pages
	API + Webhooks
	Publish, ads management, Messenger
	OAuth2
	TikTok Business
	API
	Publish, analytics
	OAuth2
	YouTube
	API
	Video upload, analytics
	OAuth2
	LinkedIn
	API
	Publish, articles
	OAuth2
	Twitter/X
	API
	Publish, replies, threads
	OAuth2
	Google Ads
	API
	Campaign creation, management, optimization
	OAuth2
	Meta Ads
	API
	Campaign creation, management, targeting
	OAuth2
	WhatsApp Business
	API
	Automated messages, chatbot, notifications
	API Key
	Telegram Bot
	Bot API
	Notifications, commands, reports
	Bot Token
	Google Analytics
	API
	Website data import
	OAuth2
	Mailchimp/SendGrid
	API
	Email campaigns, audience management
	API Key
	Stripe
	Webhooks + API
	Payments, subscriptions, invoices
	API Key
	Zapier/n8n
	Webhooks
	Connect with 1000+ apps
	Webhooks
	________________


Chapter 28: No-Code Skill Expansion
NEXUS uses a skill-based architecture. New capabilities can be added without writing application code.


28.1 How to Add a New Department (4 Steps)
66. Create Skill JSON: Define skill name, input schema, output schema, prompt template, and API connections
67. MoltGuard Security Scan: Automated scan for code safety, permissions, API access, data leaks. Blocks if suspicious.
68. Register in Database: Skill saved in PostgreSQL Skills table. Chain Manager can now route tasks to it.
69. Agent Uses It: Any agent can use the skill. Orchestrator routes tasks based on skill registry. Zero code changes. No restart needed.


28.2 Example: Adding an Accounting Agent
* Create 5 skills: invoice-gen, expense-tracker, tax-calc, report-builder, receipt-scanner
* MoltGuard approves all 5 skills
* Skills registered in database
* New 'Accounting Agent' created with these 5 skills assigned
* Agent is live immediately. No application code changed. No server restart.


28.3 Skill JSON Structure
{
  "name": "invoice-gen",
  "description": "Generate professional invoices from structured data",
  "input_schema": { "client_name": "string", "items": "array", "tax_rate": "number" },
  "output_schema": { "pdf_url": "string", "total": "number" },
  "prompt_template": "Generate a professional invoice for {client_name}...",
  "apis": ["pdf-generation"],
  "model_preference": "haiku",
  "estimated_cost": 0.01
}
________________


Chapter 29: Development Phases
Build Plan Summary
Phase
	Commands
	Description
	Sessions
	Phase 1
	Fix Issues
	Fix current UI issues, design system
	1
	Phase 2
	Missing Client Pages
	CRM, Analytics, Email, Knowledge Base
	2-3
	Phase 3
	Mission Control (Admin)
	6-tab admin dashboard
	2-3
	Phase 4
	Settings, Reports, Landing
	Settings pages, report system, public landing
	1-2
	Phase 5
	Skills & Advisor
	Skill management, NEXUS Advisor page
	1
	Phase 6
	Agent Workspaces
	Content Queue, Lead Discovery, SEO Dashboard
	2-3
	Phase 7
	Final Verification
	Test every page and feature
	1
	Phase 8A
	Backend Infrastructure
	Project setup, database, auth
	2-3
	Phase 8B
	API Endpoints
	All 60+ API endpoints
	3-5
	Phase 8C
	Connect Frontend
	Wire frontend to real APIs
	2-3
	Phase 8D
	AI Service + Skills
	Install OpenClaw skills, AI service bridge
	1-2
	Phase 8E
	Deploy & Launch
	Production build, PM2, Caddy, go live
	1
	TOTAL
	

	Complete Platform
	19-28 sessions
	________________


End of Document


NEXUS GEN3.AI - Complete Development Guide v1.0
March 2026 - Confidential - For Development Team Only
24 Agents | 101+ Skills | 14 Departments | 9 Engines | 23+ Chains | 6 Heartbeats | 3 Memory Tiers


This document contains everything the development team needs to build, understand, and maintain the NEXUS platform. Every page, every agent, every skill, every workflow, every API endpoint, and every database table is documented here.
For questions: Contact Med Talhi - Founder - gen3.ai
Page   |  gen3.ai  |  March 2026