Prism

A social forecasting platform for turning beliefs about the future into calibrated probabilities and shareable prediction-market portfolios.

Live app · Explore the model

Prism combines a multi-agent forecasting engine with live Kalshi market data. Users can research a binary question, express a broader thesis as a weighted basket of contracts, and publish that thesis for others to discover and follow.

What it does

• Agentic forecasting — parses a question, establishes independent outside-view base rates, applies inside-view updates, reconciles disagreements, and returns a calibrated probability with an evidence trail.
• AI basket construction — interviews the user about a belief, maps it across relevant domains, screens live markets, and proposes a weighted portfolio of Kalshi contracts.
• Manual basket studio — lets users search markets, select outcomes, size positions, and save a thesis without the AI workflow.
• Social research layer — supports public and private theses, creator profiles, follows, personalized feeds, bookmarks, and performance views.
• Basket trading — signed-in users can connect their own Kalshi API credentials, preview a dollar allocation, confirm limit orders, and review Prism-attributed fills and portfolio marks. Live execution requires ENABLE_LIVE_TRADING=true on the API service.

Architecture

flowchart LR
    UI[Next.js web app] --> API[FastAPI service]
    API --> Forecast[Multi-agent forecaster]
    API --> Basket[Trading companion]
    Forecast --> Router[OpenRouter]
    Basket --> Router
    Forecast --> Search[Web research]
    Basket --> Kalshi[Kalshi market data]
    API --> DB[(PostgreSQL / SQLite)]
    UI --> Auth[Supabase Auth]

The forecasting pipeline separates outside-view reasoning from inside-view reasoning. Three base-rate agents build independent reference classes, a reconciler establishes a prior, inside-view agents update it, and a supervisor produces the final ensemble estimate. Platt scaling is applied before the result is returned.

Tech stack

Layer	Technology
Web	Next.js 16, React 19, TypeScript, Framer Motion
API	FastAPI, Pydantic, Uvicorn
AI	OpenRouter, structured multi-agent workflows
Markets	Kalshi API and local market caches
Data and auth	PostgreSQL with SQLite fallback, Supabase Auth
Deployment	Vercel and Railway

Repository map

forecaster/           Core forecasting engine and CLI
trading_companion/    Belief analysis and basket-construction agents
prism/api/            FastAPI application and persistence layer
prism/frontend/       Next.js product experience
tests/                Deterministic forecasting unit tests

Run locally

Prerequisites

• Python 3.11+
• Node.js 20.9+
• An OpenRouter API key
• Kalshi credentials for live market data
• Supabase credentials for authentication features

1. Start the API

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env
uvicorn prism.api.main:app --reload --port 8000

Fill in the credentials you need in .env. Without DATABASE_URL, the API uses a local SQLite database.

2. Start the web app

cd prism/frontend
npm ci
cp .env.example .env.local
npm run dev

Open http://localhost:3000. The local API runs at http://localhost:8000.

CLI forecasting

The forecasting engine can also run independently:

python -m forecaster forecast \
  "Will the Federal Reserve cut rates before July 2027?" \
  --runs 3 \
  --output forecast.json

Configuration

Variable	Purpose
OPENROUTER_API_KEY	Model access for forecasting and basket agents
KALSHI_API_KEY	Kalshi API identifier
KALSHI_PRIVATE_KEY_FILE	Path to a Kalshi private key file
DATABASE_URL	PostgreSQL connection string; optional locally
SUPABASE_URL	Supabase project URL used by the API
SUPABASE_JWT_SECRET	Validates authenticated API requests
KALSHI_CREDENTIALS_ENCRYPTION_KEY	Fernet key used to encrypt each user's Kalshi key ID and private key at rest; keep server only and retain for existing connections
ENABLE_LIVE_TRADING	Set to true after deployment checks to allow confirmed Kalshi order submission
PRISM_ADMIN_API_KEY	Optional secret for diagnostic and maintenance endpoints; if unset, they are unavailable
ALLOWED_ORIGINS	Comma-separated CORS allowlist
NEXT_PUBLIC_API_URL	Browser-facing API base URL
NEXT_PUBLIC_SUPABASE_URL	Supabase project URL used by the frontend
NEXT_PUBLIC_SUPABASE_ANON_KEY	Supabase browser client key

See the checked-in .env.example files for the full local templates.

Quality checks

python -m unittest discover -s tests -v
python -m compileall -q forecaster trading_companion prism/api

cd prism/frontend
npm run lint
npm run build

Disclaimer

Prism is a research and portfolio-construction project. Its forecasts and market views are not financial advice.