Build on tokenized markets through one coherent interface.

Everything between intent and signature.

From wallet intent to an inspectable plan.

Initialize the client, describe the exact-input trade, and receive ordered unsigned transactions with a simulation result.

import { RailXClient } from "@railx/sdk";

const rail = new RailXClient({
  baseUrl: "https://railx.dev/api/v0",
  apiKey: process.env.RAILX_PROJECT_KEY,
});

const plan = await rail.trades.plan({
  wallet: address,
  intent: {
    type: "exact_input",
    tokenIn: "USDG",
    tokenOut: "NVDA",
    amountIn: "500",
  },
});

{
  "status": "ready",
  "chainId": 4663,
  "intent": {
    "type": "swap_exact_input",
    "tokenIn": "USDG",
    "tokenOut": "NVDA",
    "amountIn": "500"
  },
  "simulation": {
    "status": "passed",
    "mode": "sequential"
  },
  "verdict": {
    "outcome": "allow",
    "policyVersion": { "id": "policy_...", "version": 3, "schemaVersion": 1 },
    "checks": [
      { "source": "rail_safety", "code": "SIMULATION_PASSED" },
      { "source": "project_policy", "code": "POLICY_EVALUATED" }
    ]
  },
  "signing": {
    "mode": "wallet_controlled",
    "submittedByRail": false
  }
}

Authenticate server-side.

Create a test or live project key in the dashboard, pass it as a bearer token, and keep it in server environment variables. Keys are isolated by environment and never belong in a browser bundle.

Test keys run against a deterministic sandbox.

A railx_test_ key never reaches Robinhood, the chain, or Uniswap. Every call is answered from a fixed catalog with repeatable balances, so your integration tests need no provider credential and produce the same result every run. Pick an outcome by wallet address, or force one with the x-railx-scenario header.

Any other wallet behaves like zero_balance. The symbol HALT reports a trading halt. Plan and quote identifiers repeat when you send the same x-request-id. Sandbox responses carry x-railx-sandbox: on.

Every plan tells you whether to allow, review, or block it.

RailX first simulates the full sequence, then applies the active immutable policy for that project. Policies can allow or block stocks and payment tokens, and set warning or blocking thresholds for trade size and requested slippage. Every result includes structured checks a person or coding agent can explain.

Seven primitives. One response model.

All routes are versioned under /api/v0. Quantities use exact decimal strings, addresses are checksummed, and every response includes a request ID.

Usage and diagnostics are built into every request.

Authenticated responses identify the resolved environment and plan, plus minute and monthly allowance headers. The dashboard separates test and live usage, searches recent requests by ID, route, status, key prefix, error, symbol, or wallet, and exports daily usage as CSV.

Errors are designed for software.

Every failure includes a stable code, plain-language message, retryability signal, details, and request ID. Branch on the error code, never the prose.