01
Find the place
The question is matched against countries, cities and regions. No place mentioned means the whole world.
Japan
Tap one to see how the crowd answered. See all predictions →
See it run
Predictr finds where your question matters, samples people in proportion to who really lives there, and lets you watch every one of them react before the verdict lands.
Live demo: real population sampling, simulated reactions.
How it works
The question is matched against countries, cities and regions. No place mentioned means the whole world.
People are drawn in proportion to real population: big cities by metro size, the rest from towns and countryside, ages from the country's age profile.
The model answers as each person, from their age, home and the latest headlines. Watch them flip from NO to YES live.
The crowd's average becomes the forecast, split by place, age and city vs countryside, with an expert-style analyst as a cross-check.
Three engines, one app
Future values of any series (sales, prices, demand) with an 80% range. Chronos-Bolt, or an ETS fallback that always runs.
Fill in a missing column from the others (price, churn, risk) with a cross-validated score and the features that drive it. TabPFN or gradient boosting.
Chance a real-world question resolves YES, from a population-sampled crowd plus an analyst. Works with any OpenAI-compatible model.
Run it yourself
Predictr runs on your computer. Connect a free local model with Ollama or LM Studio and your files and questions stay on it (turn off the news search for fully offline use), or paste a key from OpenAI, Groq, OpenRouter, Gemini and more.
Download Predictr (.zip)start.bat.brew install python).start.command → Open (first time only).sudo apt install python3-venv)../start.sh in the folder.