AI Marketing Agency — Project Context Loader

• Opus: architecture decisions, complex synthesis, INSIGHT reports
• Sonnet: research, content creation, tool comparison, scraping
• Haiku: signal detection, classification, lightweight data extraction

3. Save Everything

Every insight from every link Pafi sends must be saved to:

• Cortex (collection: "research", metadata: {project: "ai-marketing-agency"})
• PROJECT-CARD (cross-reference section)
• Notion Intel Hub (database: 327d31d1-1b7d-819a-972a-c10f82224289) This applies regardless of topic — even if a link seems unrelated, store it.

3b. Backlink Everything to PROJECT-CARD

Every piece of work on this project MUST be backlinked to the PROJECT-CARD:

• New research → add Cortex ID + summary to PROJECT-CARD Research section
• New decisions → add to PROJECT-CARD Decisions section
• New partnerships → add to PROJECT-CARD Partnerships section
• New procedures → add to PROJECT-CARD Procedures section Path: /Users/pafi/.nexus/projects/ai-marketing-agency/PROJECT-CARD.md

3c. Undecided Items → braindump

Decisions "in discussion" but NOT yet confirmed go to braindump, not PROJECT-CARD:

• Possible future tasks
• Ideas under consideration
• Items waiting for Leo's decision
• Options being evaluated Path: /Users/pafi/.nexus/projects/ai-marketing-agency/braindump.md

3d. Notion DB Routing

Three databases, three purposes:

• Intel Hub (327d31d1-1b7d-8129-b036-000b23f88829) — LIVE MONITORING SOURCES ONLY: X accounts, YT channels, GitHub orgs, newsletters, blogs that update regularly. SourceChannel = account/channel URL (not individual post).
• Content Inbox (326d31d1-1b7d-81bd-b9f1-000bdac0f30b) — EXTRACTED CONTENT: insights from specific links, competitor analysis, article summaries, video takeaways. URL = the specific post/page.
• Master Tasks (31bd31d1-1b7d-810d-a0cf-000beeddeee1) — ACTION ITEMS: tasks to execute, deadlines, assignments.
• URL format: https://www.notion.so/{id-without-hyphens}
• Use mcp__notion__API-post-page to create pages, mcp__notion__API-query-data-source to query
• Link Notion page ID back in PROJECT-CARD for cross-reference
• Also available: Decisions Log (31bd31d1-1b7d-8156), Contacts (31bd31d1-1b7d-8114), Active Projects (31cd31d1-1b7d-817e)

3e. Intel Hub SourceChannel Rule

SourceChannel = the CHANNEL/ACCOUNT, not the individual post:

• X: https://x.com/handle (not the status URL)
• YouTube: https://youtube.com/@channel (not the video URL)
• GitHub: https://github.com/org-or-user (not the specific repo)
• Instagram: https://instagram.com/handle
• Website: root domain (e.g., https://okara.ai) The individual post/video/repo URL goes in the URL field.

4. Research Tools (COST RULES)

• USE: mcp__perplexity__search (Sonar Pro) — primary research
• USE: Gemini Deep Research via Gemini CLI — for deep dives
• DO NOT USE: mcp__perplexity__deep_research — too expensive
• USE: mcp__brave-search__brave_web_search, mcp__tavily__tavily_search — supplementary
• USE: GitHub search, Apify actors — for tools/repos/social
• ALWAYS include GitHub/repos in research scope

5. Proactive Research

When new insights arrive, automatically command additional research if something seems missing. Don't wait for Pafi to ask — identify gaps and fill them.

6. Leo Decides

The following decisions belong to Leo (co-founder, marketing guru):

• Brand name and personality
• Pricing (validated range: $149-$4,999/mo across segments)
• Product list and launch order
• Visual identity and design choices

7. Verification

Periodically check if background agents are blocked or zombie. Restart as needed.

Business Decisions (CONFIRMED)

Identity

• Market: Global + Romania + Italy
• Timeline: AGGRESSIVE — 30 days to first client
• Co-founders: Pafi + Leo + possibly Cody
• Legal: Delaware LLC (existing, keep) + Romania SRL (add, 1% CIT sub 100K EUR)
• Budget: Max $1,000/mo bootstrap, prefer build over buy
• First clients: Warm market
• Partnership: Luigi Emanuele Foscale / AI.ANIMA (ai-anima.ai)
  • Grownnectia = 1,400+ startups = biggest client pipeline
  • 14 businesses, serial entrepreneur, 46K IG followers

Segments (ALL 4 from start)

1. B2B — Small ($199/mo) / Growth ($599/mo) / Enterprise ($1,999+/mo)
2. B2C/KOL — Creator ($149/mo) / KOL Pro ($399/mo) / KOL Enterprise ($999+/mo)
3. B2P Political — Local ($499/mo) / Regional ($1,499/mo) / National ($4,999+/mo)
4. Products — Standalone (AI Audit, Lead Finder, Content Waterfall, Brand Voice Kit, etc.)

Pricing Gap (VALIDATED)

Target: $499-$1,499/mo "intelligent execution" band

• Below: SaaS tools $19-499 (commodity, shallow)
• Above: Agencies $2,500-$25K+ (premium, fully booked)
• Nobody does full-funnel autonomous execution at $500-1,500/mo

Support Model

• B2C: AI complet (possibly via AI.ANIMA Digital Humans), human only if AI fails
• B2B/Enterprise/B2P: AI does filtering + lead prep, human takes over for close

Technical Decisions (CONFIRMED)

Architecture

• Orchestration: LangGraph (scored 82/90, #1 recommendation)
• Frontend: Next.js 16 + React 19 + Tailwind + shadcn/ui + Framer Motion
• Backend: FastAPI (Python, async-first)
• AI Gateway: LiteLLM or Portkey (multi-model routing + circuit breakers)
• Database: Supabase (PostgreSQL + RLS for multi-tenancy)
• Vector DB: Qdrant (separate instance from NexusOS)
• Hosting: Vercel (frontend) + Railway (backend) + Cloudflare (CDN)
• Auth: Clerk (fast MVP) or better-auth (ownership)
• Payments: Stripe Billing + Metronome (AI token metering)
• CRM: GoHighLevel hybrid — GHL for infrastructure + custom AI layer on top
• Social Publishing: Late/Zernio $49/mo unified API (NOT build direct)
• Monitoring: Langfuse + Helicone + Sentry + OpenTelemetry

Multi-Model Routing (saves 50-70% costs)

• Claude Sonnet → copy, strategy, nuance
• Claude Opus → code generation, complex orchestration
• GPT-5.2 → structured JSON, math, function calling
• Gemini 3 Flash → bulk content (6-8x cheaper)
• Gemini 3 Pro → long docs (1M context)

Agent Architecture (11 + orchestrator)

Each agent has: defined model, explicit boundaries (does NOT), signal tiering, error handling. Content waterfall: Research → Strategy → Write → SEO/Social/Email → Analytics → Auto-Optimization.

Research Cortex IDs (18 completed)

Additional Intel Cortex IDs

Key Competitive Insights

1. GEO (Generative Engine Optimization) = new SEO — track brand visibility in AI chatbots
2. CORD governance (ALLOW/CONTAIN/CHALLENGE/BLOCK) — best agent safety framework
3. HAAO — 88 agents per human (Moltbook/Meta benchmark)
4. Single Grain FULLY BOOKED at $25K+/mo — requires Claude Code for ALL staff
5. Content waterfall = universal pattern (Research→Strategy→Write→SEO→Social→Outreach)
6. EU political ads: Meta/Google/TikTok EXITED — only X remains
7. Brand Voice Engine = #1 differentiator — nobody does feedback loop + constitutional review

Satellite View

ACUM ──> FUNDATIE ──> MVP ──> VALIDARE ──> SCALE ──> ENTERPRISE
 |         |           |        |            |          |
 0 cli.    0 cli.      5 cli.   50 cli.      200 cli.   1000 cli.
 $0        $0          $1K MRR  $15K MRR     $60K MRR   $300K MRR
 |         |           |        |            |          |
 Research  Brand+Web   B2B      +B2C/KOL     +Ads+SEO   +B2P+WL
 Design    Onboarding  Starter  +Growth      +Products  +Marketplace
 Agents    4 agents    Pipeline Full waterf. Full stack  Governance

Remaining Questions for Leo

1. Brand name (generăm opțiuni, Leo alege)
2. Brand personality (dual-tone premium B2B + accessible B2C?)
3. Pricing final per segment/tier
4. Product list and launch order
5. Cody — co-founder da/nu?