Startup Agent OS

Multi-agent infrastructure to run a startup from idea to scale. Voice-first AI operating system.

Architecture

Idea → Voice Dump → Idea Evaluator → [Ops Agent, Product Agent, Sales & Marketing Agent]

3 agents, 41 skills covering every function of a startup:

Ops Agent (7 skills)

Strategy, business analysis, and operational planning.

• Executive Summary (problem breakdown, why-now, market trends, risks, exit strategy)
• Competitive Analysis & Differentiation
• Business Model & Monetization
• ICP / Personas / Audience Behavior
• Positioning
• PRD (Product Requirements Document)
• Solution & GTM (Go-to-Market)

Product Agent (10 skills)

Takes a PRD and turns it into a working product.

• Engineering (architecture, stack selection, API design)
• Backend Code
• Frontend Code
• Prompt Engineering
• QA (test plans, automated tests)
• Design (design system, UX principles)
• Wireframes
• Asset Manager
• Website Architecture
• MVP Scoping

Sales & Marketing Agent (24 skills)

All go-to-market execution.

• Content Engine (video, written, image)
• Lead Generation & Enrichment
• Outreach (paid ads, email campaigns, Product Hunt)
• Community (events, education, partnerships)
• Customer Success (signup flow, landing pages, launch posts)
• Analysis (A/B testing, feedback analysis)
• Content Scheduling
• Documentation

How It Works

1. Voice Dump — Speak or type your idea
2. Idea Evaluator — AI evaluates feasibility, novelty, and market fit
3. Delegation — Tasks are routed to the right agent and skill
4. Execution — Each skill produces a specific deliverable
5. Memory Bus — All context is shared across agents for coherence

Quick Start

# Install
npm install

# Set your API key
export ANTHROPIC_API_KEY=sk-ant-...

# Run the voice terminal (text mode)
npm start

# Or run a specific skill directly
npx tsx src/index.ts

Commands

/idea <text>     — Evaluate a new startup idea
/run <skill-id>  — Run a specific skill (e.g., /run ops:prd)
/agents          — List all agents and skills
/memory          — Show shared context
/quit            — Exit

Development

# Type check
npm run lint

# Run tests
npm test

# Watch mode
npm run dev

Tech Stack

• TypeScript — Strict mode, no any
• Claude API (Anthropic SDK) — LLM backbone
• Pino — Structured logging
• Zod — Schema validation
• Vitest — Testing

Voice Integration

The Voice Terminal supports pluggable transcription (Whisper API, Deepgram) and synthesis (OpenAI TTS, ElevenLabs). In text mode by default — voice mode integrates with clui-cc (Voice Terminal Electron app).

License

MIT