_y

# _y Holdings — Your AI Company That Never Sleeps

[![CI](https://github.com/antryu2b/_y/actions/workflows/ci.yml/badge.svg)](https://github.com/antryu2b/_y/actions)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![Tests](https://img.shields.io/badge/tests-407_passing-brightgreen.svg)](./__tests__)
[![Agents](https://img.shields.io/badge/agents-30-purple.svg)](./src/data/agent-config.ts)
> 30 AI agents running your company 24/7. Setup in 5 minutes. No credit card required.
> See what your AI agents are actually doing.
>
> Independent analysis. Visual tracking. Structured synthesis. Local-first.

![Visual AI Agent Orchestration Dashboard — _y](./docs/demo.gif)

## What is this?

_y is a visual orchestration layer for AI agents. Most multi-agent frameworks run in a terminal — logs scroll, JSON outputs, you have no idea what's happening between agents.

_y makes agent work **visible**. See who's analyzing what, how reports flow between functions, where agents disagree, and which LLM provider catches what the others miss. Configure agents for your business functions — marketing, engineering, risk, finance — and watch them work independently before a synthesis agent combines everything.

Run locally with **Ollama** (free), or use cloud LLMs (OpenAI, Anthropic, Google). Mix different providers per function for broader coverage.

## What You Get

Connect your business URL and _y's agents go to work:

### 📊 Strategic Reports
Each department analyzes your business independently:

```
┌─────────────────────────────────────────────────────────┐
│ REPORT: Market Positioning Analysis                     │
│ Agent: Searchy (5F Marketing)                           │
│ Model: gemini-2.0-flash                                 │
├─────────────────────────────────────────────────────────┤
│                                                         │
│ Finding: Target site ranks #47 for primary keyword      │
│ "AI automation" — competitors hold positions #3-#12.    │
│                                                         │
│ Recommendation: Focus on long-tail keywords             │
│ "AI company builder" and "local LLM agents" where       │
│ competition is 10x lower.                               │
│                                                         │
│ Risk: Skepty (8F Risk) flags keyword cannibalization    │
│ between blog and product pages.                         │
│                                                         │
│ Status: PENDING REVIEW → Chairman Dashboard             │
└─────────────────────────────────────────────────────────┘
```

### 🏛️ Decision Pipeline
Reports flow through a structured chain — not a chatbot:

```
URL Input → Agent Analysis (independent, parallel)
         → Cross-Department Review
         → Skepty Challenge (independent oversight)
         → Counsely Synthesis (Chief of Staff)
         → Chairman Decision (you)
```

### 🔄 What the agents actually do

| Agent | Department | Example Output |
|-------|-----------|----------------|
| **Searchy** | Marketing | SEO audit, competitor keyword gaps |
| **Buildy** | Engineering | Tech stack analysis, performance bottlenecks |
| **Finy** | Capital | Revenue model assessment, unit economics |
| **Skepty** | Risk | Flags blind spots in other agents' reports |
| **Buzzy** | Content | Content strategy, social media positioning |
| **Counsely** | Chairman Office | Synthesizes all reports into executive brief |

> **Key:** No agent sees another's analysis until review phase. This prevents groupthink — the Byzantine Principle in practice.



## Quick Start

```bash
# 1. Clone
git clone https://github.com/antryu2b/_y.git
cd _y

# 2. Install
npm install

# 3. Setup (auto-detects your hardware, recommends models)
npm run setup

# 4. Start
npm run dev

# 5. Start the chat worker (in another terminal)
npm run chat-worker
```

Open [http://localhost:3000](http://localhost:3000) and connect your company.

## After Setup — What to Do

Once you see the dashboard at `localhost:3000`:

### Step 1: Enter a business URL
Type any company website URL into the input field. The agents will analyze it.

### Step 2: Watch agents work
Each agent independently analyzes the URL from their department's perspective:
- **Searchy** checks SEO and search positioning
- **Buildy** audits the tech stack
- **Finy** evaluates the business model
- **Skepty** challenges what others might miss

### Step 3: Read the reports
Reports appear in the **Reports** panel. Each department submits independently — no agent sees another's work until synthesis.

### Step 4: Review the synthesis
**Counsely** (Chief of Staff) combines all department reports into one executive brief with recommendations.

### Step 5: Make decisions
Items flow to the **Decision Pipeline** where you approve, reject, or modify recommendations.

### Example workflow
```
You enter: https://example-startup.com

→ Searchy: "SEO score 34/100, missing meta descriptions on 12 pages"
→ Buildy: "React 18, no SSR, 4.2s load time on mobile"
→ Finy: "Freemium model, estimated 2.3% conversion rate"
→ Skepty: "Buildy missed: third-party scripts blocking render"
→ Counsely: "Priority: fix mobile performance (affects 68% of traffic)"

→ You: Approve / Modify / Reject
```

> **Pro tip:** Try your own company's URL first. Then try a competitor's.

## Hardware-Aware Setup

The setup wizard detects your RAM/GPU and recommends the optimal model profile:

| Profile | RAM | Models | Download |
|---------|-----|--------|----------|
| SMALL | 8GB | qwen2.5:7b | ~4GB |
| MEDIUM | 16GB | qwen3:14b + gemma3:12b | ~20GB |
| LARGE | 32GB | qwen3:32b + gemma3:27b | ~55GB |
| X-LARGE | 64GB+ | + llama3.3:70b | ~97GB |

The setup automatically pulls Ollama models and generates a `llm-profile.json` for optimal agent-model matching.

## LLM Providers

Choose your AI backend during setup:

| Provider | Models | Cost | Requirements |
|----------|--------|------|-------------|
| **Ollama** (default) | Qwen3, Gemma3, Llama3, ExaOne | Free | 8GB+ RAM, Ollama installed |
| **OpenAI** | GPT-4o, GPT-4o-mini | Pay per token | API key |
| **Anthropic** | Claude Sonnet, Claude Opus | Pay per token | API key |
| **Google** | Gemini Flash, Gemini Pro | Free tier available | API key |
| **Mixed** ⭐ | Any combination above | Varies | Multiple keys |

**Mixed mode** is where _y shines — assign different providers to different departments:

```env
# .env
LLM_PROVIDER=mixed
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GOOGLE_API_KEY=AI...
```

```json
// llm-profile.json (auto-generated by setup)
{
  "provider": "mixed",
  "agents": {
    "counsely": { "provider": "anthropic", "model": "claude-sonnet-4-20250514" },
    "skepty": { "provider": "openai", "model": "gpt-4o" },
    "searchy": { "provider": "google", "model": "gemini-2.0-flash" },
    "buildy": { "provider": "ollama", "model": "qwen3:32b" }
  }
}
```

Byzantine Principle in action: analysis (Gemini) → challenge (GPT-4o) → synthesis (Claude). Different companies, different architectures, different blind spots.

## Database

_y supports three database backends:

### SQLite (Default)
Zero configuration. Data stored locally in `data/y-company.db`.

```bash
# No setup needed — tables auto-created on first run
```

### PostgreSQL
For production deployments with multiple users.

```bash
# Set in .env:
DB_PROVIDER=postgres
DATABASE_URL=postgresql://user:password@localhost:5432/y_company

# Create tables:
psql $DATABASE_URL < sql/postgres-schema.sql
```

### Supabase
Cloud PostgreSQL with authentication and realtime features.

```bash
# Set in .env:
DB_PROVIDER=supabase
SUPABASE_URL=https://your-project.supabase.co
SUPABASE_ANON_KEY=your-anon-key
SUPABASE_SERVICE_KEY=your-service-role-key

# Create tables in Supabase SQL Editor:
# Copy contents of sql/postgres-schema.sql
```

### Storage Location (SQLite)
```
data/y-company.db
```

### Tables (auto-created)
| Table | Purpose |
|-------|--

[truncated…]