BetterAI vs
SimpleFunctions.

BetterAI runs an ensemble of LLMs against a specific event and returns a single explainable analysis. SF's thesis system does something structurally different: you submit a sentence like 'US avoids recession in 2026,' and SF decomposes it into a causal tree of testable sub-claims, maps each sub-claim to matching contracts on Kalshi and Polymarket, propagates probabilities through the tree, and then runs a continuous evaluation heartbeat — news scan, price refresh, milestone check, LLM eval, confidence update — until you close the thesis. It is a persistent reasoning loop, not a one-shot analysis.

Portfolio Autopilot is SF's autonomous trading agent. Before any order is placed, a position must clear seven gates in sequence: a manual kill switch, maximum position size limits, a max drawdown gate, a regime check (adverse-selection classification), a liquidity gate via the LAS indicator, a conviction threshold from the thesis evaluation, and a cross-venue price sanity check. The decision cycle uses a 1M-context LLM fed by 13 data sources including live orderbook, calibration history, and thesis signals. No gate can be bypassed programmatically.

Yes. Run claude mcp add simplefunctions --url https://simplefunctions.dev/api/mcp/mcp and 56 tools become available in any MCP-compatible client, including Claude Code and Cursor. Tools cover price queries, orderbook depth, cross-venue arbitrage pairs, thesis creation, signal injection, trade ideas, world snapshot, and calibration data. BetterAI does not publish an MCP server.

Based on publicly available sources, BetterAI is described as a prediction platform delivering real-time, explainable analyses through a product interface. No public REST or programmatic API has been documented in sources available as of 2026-04. If you need a machine-readable API for agent integration, SimpleFunctions exposes a full REST API with no auth required for reads, an OpenAPI spec at /openapi.json, and an MCP server at /api/mcp/mcp.

SF computes rolling Brier scores for its own price estimates and publishes them at /api/calibration, broken down by venue, event category, and price bucket, over a 90-day trailing window. Current figures: Kalshi 0.20, Polymarket 0.12 on T-24h price. You can re-verify these numbers yourself with a single curl call. Most prediction market platforms claim accuracy without a public audit trail; SF makes the raw scoring data queryable.

SF pre-computes six indicators across 48K+ active contracts: IY (implied yield, annualised return if the contract resolves YES), CRI (cliff risk index, measuring how sharply price drops near expiry), LAS (liquidity availability score, bid/ask spread and depth composite), EE (event overround, total probability excess in multi-outcome markets), τ-days (time to settlement), and regime label (adverse-selection classification per contract). All are available at /screen for screening and at the individual market endpoint.

GET /api/agent/world returns approximately 800 tokens covering the current state of the most significant open contracts across both venues — prices, volume, recent moves, active thesis signals, and top trade ideas. A delta endpoint returns only what changed since the last call. This lets an AI agent maintain a compact but current world model without polling dozens of individual endpoints. The endpoint is designed to fit inside an agent's context window as a standing briefing, refreshable on each reasoning cycle.