"If we can't kill it, the market won't."
An agentic AI tool that stress-tests your startup idea before you waste months building it. It's not a cheerleader — it's an adversarial validator. Enter your idea, and KillMyIdea researches the web to find real competitors, analyzes them deeply, and gives you a brutally honest KILL or SURVIVE verdict. If your idea survives, it helps you build it.
- Validator — rejects gibberish/nonsense before wasting any API calls
- Agent 0: Analyst — ethics check, extracts keywords, finds top 5 real competitors via live web search
- Agents 1–5: Researchers — each digs deep into one competitor (funding, product, traction, news)
- Agent 6: Judge — compares your idea against all 5 and delivers a structured KILL or SURVIVE verdict
- Agent 7: Builder (only if you survive) — MVP scope, tech stack recommendation, and GTM plan
[Your Idea]
│
▼
[Validator] → rejects nonsense early
│
▼
[Agent 0: Analyst] → ethics check + find top 5 competitors
│
├── [Agent 1: Researcher] → Competitor A
├── [Agent 2: Researcher] → Competitor B
├── [Agent 3: Researcher] → Competitor C
├── [Agent 4: Researcher] → Competitor D
└── [Agent 5: Researcher] → Competitor E
│
▼
[Agent 6: Judge] → KILL or SURVIVE verdict
│
[If SURVIVE]
▼
[Agent 7: Builder] → MVP plan, stack, roadmap
| Layer | Tool | Purpose |
|---|---|---|
| Agent Framework | LangGraph | Multi-agent state graph |
| LLM | Google Gemini | Reasoning + analysis |
| Web Search | Tavily API | Live competitor research |
| Short-term Memory | LangGraph state | Chat history per session |
| Long-term Memory | ChromaDB | Store company sheets for RAG |
| Backend | FastAPI + Uvicorn | REST API |
| Frontend | Next.js | Interactive UI |
- Node.js 18+
- Python 3.10+
- Google Gemini API key
- Tavily API key
# Install all dependencies (frontend + backend)
npm run install
# Copy env template
npm run setupThen edit backend/.env with your keys:
GOOGLE_API_KEY=your_gemini_key_here
TAVILY_API_KEY=your_tavily_key_herenpm run dev- Frontend → http://localhost:3000
- Backend API → http://localhost:8000
Color-coded terminal output — yellow for backend, cyan for frontend. Ctrl+C stops both.
Each agent has its own model, swappable via backend/.env — no code changes needed:
ETHICS_MODEL=gemini-2.0-flash-lite
ANALYST_MODEL=gemini-2.0-flash
RESEARCHER_MODEL=gemini-2.0-flash-lite
JUDGE_MODEL=gemini-2.5-pro
BUILDER_MODEL=gemini-2.5-proKillMyIdea/
├── backend/
│ ├── agents/
│ │ ├── validator.py # pre-flight idea validity check
│ │ ├── analyst.py # orchestrator + competitor finder
│ │ ├── competitor_researcher.py # reused for all 5 competitors
│ │ ├── judge.py # KILL or SURVIVE verdict
│ │ └── builder.py # MVP + GTM plan
│ ├── tools/
│ │ ├── web_search.py # Tavily integration
│ │ ├── crunchbase.py # funding data
│ │ ├── scraper.py # homepage scraping
│ │ └── memory.py # ChromaDB read/write
│ ├── graph/
│ │ └── pipeline.py # LangGraph state graph
│ ├── config.py # per-agent model config
│ └── main.py # FastAPI entry point
├── frontend/
│ └── app/ # Next.js UI
├── package.json # root scripts
└── .env.example # template for API keys
MIT