Research with ChatGPT and Codex

Screen U.S.-traded assets by supported exposure, structure, leverage, expenses, and available price history. Missing metadata stays visible.

Build a private allocation model, inspect the actual code and backtest, and test input variations while keeping the approved strategy unchanged.

Compare saved runs over a common period. Inspect code, inputs, dependencies, backfill assumptions, data dates, and current target allocations.

The initial U.S. public directory release is being prepared. A verified installation link will appear here after publication.

Directory installation will connect to the existing ENSEMBLE service using OAuth. Sign in to your intended ENSEMBLE account, or use the existing account registration flow when prompted. Your research remains available through the website and other supported clients.

For custom-server development connections in supported host settings, the Streamable HTTP endpoint is https://api.ensemble.markets/mcp. Custom connections and directory installation are separate paths; available host controls depend on your account and workspace settings.

Builds and runs use your existing account credits, metered usage, entitlements, and spending limits. Projected cost is an admission estimate; completed work reports its actual usage. A refused request does not start a job.

Long-running work returns an identifier. If it is still pending, you can return later to inspect that same model or run without restarting it.

Coverage varies by asset and date. Backfilled pre-inception history represents explicit assumptions, not observed returns. Backtests are hypothetical, can omit real-world costs, and do not guarantee future performance.

ENSEMBLE is research software by NSMBL.IO. It does not place brokerage trades, transfer money, provide individualized investment advice, or establish whether an asset is tradable in your brokerage account.