ideakiller 0.1.0

Destroy your startup idea before the market does.

IdeaKiller

Your startup idea destroyed in 60 seconds.

IdeaKiller runs your business idea through 7 adversarial lenses and returns a Survival Score (0–100) with evidence-backed findings. Built for founders who want brutal honesty before they burn runway.

---

Features

• 7 adversarial lenses — market timing, unit economics, competition, customer acquisition, team risk, regulatory exposure, and technology feasibility
• Weighted geometric mean scoring — a fatal flaw in any single lens tanks the overall score, as it should
• Local-first LLM — uses Ollama by default; falls back to Anthropic API automatically
• Three interfaces — CLI for scripts, REST API for integrations, Gradio UI for humans
• Structured JSON output — machine-readable results for downstream tooling
• Docker-ready — single-command deployment with health checks included

---

Survival Score

Score = exp(Σ weight_i × ln(survival_probability_i)) × 100

Score	Verdict
0–20	DEAD ON ARRIVAL
21–40	CRITICAL
41–60	HIGH RISK
61–80	VIABLE
81–100	STRONG

Lens Weights

Lens	Weight	What it hunts
Market Timing	20%	Too early? Too late? Macro headwinds?
Unit Economics	20%	CAC/LTV death traps, margin collapse
Competition	15%	Incumbent moats, funded clones
Customer Acquisition	15%	Reachable customers, GTM viability
Team Risk	10%	Skills gap, key-person risk
Regulatory	10%	FDA, HIPAA, financial regs, IP landmines
Technology	10%	Unproven tech, feasibility gaps

---

Quick Start

Install:

pip install -e .

Configure LLM backend (pick one):

# Option A — local inference via Ollama (recommended)
ollama pull llama3.2

# Option B — Anthropic API
export ANTHROPIC_API_KEY=sk-ant-...

Run:

# Plain-text analysis
ideakiller analyze "Uber for dog walking with AI routing"

# JSON output
ideakiller analyze --output json "Blockchain supply chain for farms"

# REST API server  →  http://localhost:8000
ideakiller serve

# Gradio web UI    →  http://localhost:7860
ideakiller ui

REST API:

curl -X POST http://localhost:8000/analyze \
  -H "Content-Type: application/json" \
  -d '{"idea": "AI therapist for Gen Z", "context": "mobile-first, US market"}'

Docker:

docker build -t ideakiller .
docker run -p 8000:8000 -e ANTHROPIC_API_KEY=$ANTHROPIC_API_KEY ideakiller

---

Architecture

ideakiller/
├── llm.py        # LLMClient — Ollama-first, Anthropic fallback
├── analyzer.py   # IdeaAnalyzer — 7-lens prompt engine, JSON extraction
├── scorer.py     # IdeaScorer — weighted geometric mean, verdict mapping
├── cli.py        # Click CLI — analyze / serve / ui commands
├── api.py        # FastAPI — POST /analyze, GET /health
└── app.py        # Gradio UI — form inputs, markdown output, examples

Request flow:

User input
  → IdeaAnalyzer  (7 × LLM calls, structured JSON per lens)
  → IdeaScorer    (geometric mean → 0-100 score → verdict)
  → Output        (CLI text | JSON | HTTP response | Gradio markdown)

LLM strategy: LLMClient probes Ollama at localhost:11434 on first call. If unavailable, it lazy-imports the Anthropic SDK and uses ANTHROPIC_API_KEY. No configuration required beyond setting the env var.

---

Development

pip install -e ".[dev]"
pytest -v
ruff check src/ tests/
bandit -r src/ -ll

---

Contributing

See CONTRIBUTING.md for guidelines on issues, pull requests, and coding standards.

---

License

MIT

---

Built by TechKnowMad Labs