Vanessa picks the best model per task — Claude for code, Perplexity for live search, Gemini for vision, Grok for current events, Ollama for privacy.