Octagon AI vs
SimpleFunctions.

SF's thesis system accepts any natural-language sentence via POST /api/thesis/create, decomposes it into a directed acyclic graph of testable sub-claims, propagates probabilities through the tree, scans Kalshi and Polymarket for contracts that map to each node, and runs a continuous evaluation heartbeat — news scan, price refresh, milestone check, LLM evaluation, confidence update. You inject signals via /api/thesis/{id}/signal and fork public theses. Octagon AI performs its own internal research and surfaces a finished forecast; users do not author or control the decomposition logic.

Yes. SF exposes 56 tools via the Model Context Protocol. Add it to any MCP client with: claude mcp add simplefunctions --url https://simplefunctions.dev/api/mcp/mcp. It works with Claude Code, Cursor, and any other MCP-compatible host. The tools cover market search, thesis authorship, orderbook depth, cross-venue arbitrage pairs, computed indicators, world snapshot, and more. Octagon AI does not publish an MCP server.

Portfolio Autopilot is SF's autonomous trading agent. Before executing any position, it passes a 7-gate risk cascade: kill switch check, position limits, drawdown gate, regime classification, liquidity availability score, conviction threshold, and correlation guard. It runs on a 1M-context LLM with 13 live data sources feeding the decision loop. It is distinct from the thesis system — Autopilot manages execution, while the thesis system manages belief propagation. Octagon AI does not offer an execution or portfolio management layer.

GET /api/calibration returns SF's own Brier scores, computed against resolved contracts over the past 90 days, broken down by venue (Kalshi, Polymarket), category, and price bucket. Current figures: Kalshi 0.20, Polymarket 0.12 on T-24h price. This is SF auditing its own accuracy publicly — you can re-run the query with curl at any time to get a fresh result. Most forecasting products, including Octagon AI, do not publish equivalent machine-readable calibration data.

Yes. GET /api/public/market/{ticker}?depth=true returns the full bid/ask ladder, spread, and slippage estimate for any contract on Kalshi or Polymarket. This is in addition to normalised prices, computed indicators, and historical data. Cross-venue arbitrage opportunities are available at /api/public/cross-venue/pairs?preset=arb. Octagon AI does not expose live orderbook data — its product layer is forecast outputs, not market microstructure.

If you are building an agent that needs to reason about prediction markets — not just receive a finished forecast — SF is the more appropriate substrate. The 56-tool MCP server integrates with Claude Code or Cursor in one command. The thesis system gives the agent a structured causal model to work from. The Autopilot handles execution. Octagon AI is optimised for teams that want to consume AI-generated forecasts as a finished output, not teams that want to build the inference and execution layer themselves.

SF pre-computes six indicators across all 48K+ active contracts: IY (implied yield — annualised return to settlement), CRI (cliff risk index — probability mass concentrated near expiry), LAS (liquidity availability score — depth-adjusted fill quality), EE (event overround — total probability excess above 1.0), τ-days (time to settlement), and regime label (adverse-selection classification). These are queryable at /screen and via MCP. Octagon AI surfaces forecast probabilities; it does not publish tradeable signal metrics derived from market microstructure.

Based on publicly available information, Octagon AI operates as a managed deep-research forecasting service. No public REST API documentation has been published as of 2026-04. If you need a programmable API for prediction market data, thesis authorship, and agent orchestration, SF's REST API, CLI, and MCP server are openly documented and require no authentication for read access.