They analyze data, monitor competitors, find problems you'd miss, and get smarter every day — all built with Claude Code.

Every morning at 7am, this lands in Slack:

Revenue. Conversion funnels. Competitor changes. Per-location anomalies.

Not a dashboard you check. A briefing that comes to you.

한국어 README · Roadmap

They talk to each other. CRO asks Psychology: "Why are users dropping off here?"

They call tools on their own. If they need more data, they query GA4, crawl competitor sites, or search the knowledge base — without being told.

Daily Slack Briefing	Self-Improvement Dashboard

Role Manager (5 AI Specialists)	GA4 Analytics

Most AI tools wait for commands. Garnet doesn't.

• Runs 24/7 — 5 AI specialists analyze your data every cycle, automatically
• Learns your judgment — Give feedback (👍 Noted / ❌ Pass), Garnet learns what matters to you
• Finds what you'd miss — Detects anomalies across all locations, channels, and funnels daily
• Portable — Switch companies with one config file. Your AI advisor follows your career

Every 30 minutes:
  Scanner → collects GA4, SNS, competitor data
  ↓
  5 Sub-Reasoners (parallel):
    📊 Data Analyst — finds patterns and anomalies
    🎬 Content Strategist — proposes content ideas with rationale
    🧭 Marketing Strategist — market positioning and growth strategy
    💰 CRO Expert — conversion bottlenecks and quick wins
    🧠 Marketing Psychologist — behavioral insights and cognitive biases
  ↓
  Reasoner → synthesizes into actionable insights
  ↓
  Advisor Inbox → You decide: 👍 Noted / ❌ Pass + optional feedback
  ↓
  Garnet learns → next cycle is more accurate

Sub-Reasoners don't just analyze pre-collected data — they actively call tools when they need more information:

• GA4 queries, funnel analysis, per-location breakdowns
• Knowledge Store semantic search (400+ learned insights)
• Competitor website crawling via Playwright
• Instagram post/account analytics
• Web search for real-time trends
• ask_expert — CRO specialist asks Psychology specialist: "What's causing this drop-off?"

Garnet doesn't tell you what to do. It tells you what's happening and suggests what to consider.

• 👍 Noted — "Good point, I'll consider this" → Garnet learns this direction is right
• ❌ Pass + text — "Not now, focusing on B2B" → Garnet learns your priorities
• No execution tracking — Garnet advises, you decide and execute

No cloud subscription. No monthly fees. Your data stays on your machine.

Garnet runs locally on:
  Mac Mini (recommended) — always-on, low power, runs 24/7 quietly
  MacBook — works fine, just needs to be running
  Any machine with Node.js 18+ — Linux, Windows (WSL)

What you need:
  Free Gemini API key (ai.google.dev)
  GA4 service account (if you want analytics)
  Slack webhook (if you want morning briefings)
  That's it. No paid subscriptions.

Engine (domain-agnostic) → never changes
Config (per-company)     → swap when you move
Knowledge (learned)      → accumulates over time

Switch companies:
1. Write config/company.md (your new business context)
2. Run bootstrap → auto-generates domain.yaml + tools.yaml
3. Start → Garnet begins learning your new domain

• 18 watch keywords monitoring competitors and trends
• Auto-collected from web/news every 2 hours
• AI-tagged by relevance and urgency (CRITICAL alerts)
• Tech Radar scans GitHub Trending daily for applicable tools

• Knowledge Store — 400+ business insights with embedding-based semantic search (LightRAG pattern)
• Episodic Memory — 1,500+ decision records for pattern matching
• Failure Registry — Time-weighted avoidance rules

Every morning at 7am:

• Revenue, purchasers, conversion rate, new vs returning
• 6-stage purchase funnel with drop-off analysis
• Per-location revenue breakdown
• AI-generated insights and recommendations

Edit config/company.md:

---
name: "Your Company"
industry: "ecommerce"
---

# Business Context

We sell handmade candles online.
Main channels: Instagram + Google Ads.
KPIs: revenue, conversion rate, CAC, retention.
Current challenge: conversion rate stuck at 2%.

That's it. Write in plain language. Garnet reads this and configures itself.

Add to .env:

GEMINI_API_KEY=your_key          # Free at ai.google.dev
GA4_PROPERTY_ID=123456789        # Your GA4 property
GA4_CLIENT_EMAIL=...             # GA4 service account
GA4_PRIVATE_KEY=...              # GA4 service account key
SLACK_WEBHOOK_URL=...            # For daily briefings

No GA4? Garnet still works — it just won't have analytics data. It can still monitor competitors and provide strategic insights.

git clone https://github.com/mark02252/garnet-ai.git
cd garnet-ai
npm install
npx prisma db push
npm run dev

First Slack briefing arrives at 7am next morning.

Day 1:    Garnet starts collecting data, learning your business
Day 3:    Insights start getting specific to your situation
Week 2:   Knowledge store has 50+ learned patterns
Month 1:  Garnet knows your priorities, gives relevant advice
Month 3:  400+ insights, tailored to how you think

SaaS:        Track MRR, churn, trial conversion
E-commerce:  Track revenue, cart abandonment, channel ROI
Hospitality: Track bookings, per-location performance
Agency:      Track client campaigns, deliverables
Any:         If you have GA4, Garnet can analyze it

┌──────────────────────────────────────────────────────────────┐
│  Agent Loop (50+ modules)                                     │
│                                                               │
│  Scanner → 5 Sub-Reasoners (with Tool Harness) → Reasoner    │
│                ↕ ask_expert (A2A)                              │
│            Reflective Critic → Advisor Inbox → Feedback Loop  │
│                                                               │
│  World Model (config-driven, domain-portable)                │
│                                                               │
│  Cycles: 30min urgency / 1hr routine / 7am briefing          │
│          6pm evening / Mon 9am weekly review                  │
└──────────────────────────────────────────────────────────────┘

Garnet automatically verifies its own insights against real data — no human labeling required.

Every cycle:
  1. Garnet generates insights with testable predictions
  2. After 24-168 hours, compares predictions against actual data
  3. Correct → confidence +0.08 (max 0.95)
     Wrong  → confidence -0.08 (min 0.10)
  4. Knowledge Store evolves without human intervention

Safe domains (auto-learn):   analytics, competitive, retention, marketing...
Manual domains (human only):  pricing, finance, paid advertising

No runaway learning — confidence moves ±0.08 per verification, capped at [0.10, 0.95]. The system learns what works, forgets what doesn't, and never touches pricing or budget decisions without you.

• Runtime: Next.js (App Router, TypeScript)
• LLM: Gemini 2.5 Flash-Lite (primary) → Gemma4 local (fallback) — free tier only
• Embeddings: Ollama nomic-embed-text (local)
• Database: PostgreSQL (Supabase) + Prisma
• Tool Harness: Cache + whitelist + sliding window rate limit + observability
• Function Calling: Gemini/Groq native tool-use, Gemma4 JSON fallback
• Analytics: GA4 Data API + Admin API
• Notifications: Slack Webhook + Telegram Bot API
• Competitor Monitoring: Playwright headless browser
• MCP: 28 preset connections (expandable)

Garnet was built by a solo marketer using Claude Code. Every line of code was written by AI, directed by a marketer who knew what needed to exist but not how to build it.

The entire system — 50+ modules, 12 registered tools, 8 evolution phases — was built through natural language conversations with Claude Code.

If you're a marketer who wants to build your own AI advisor, you can. Start here →

MIT

Jung Jaeho — Solo marketer who built this with Claude Code.

• LinkedIn
• Portfolio

Questions, feedback, or collaboration? Open an Issue or reach out on LinkedIn.

• Roadmap
• Agent Workflows