Rate limits

• ip — per source IP (behind the ingress; respects X-Forwarded-For set by the trusted load balancer). Cheap edge defence that applies to anonymous and authenticated traffic alike. An IPv6 caller is counted per /64 prefix, not per address, so every address in one /64 shares a bucket. IPv4 is counted per address.
• wallet — per your key’s associatedWallet (single_wallet partners) or per the X-User-Wallet header value (multi_wallet partners). Prevents a single wallet identity from exceeding the platform’s per-user ceiling regardless of how the calls are distributed across your keys.
• partner — per partner, shared by all of your API keys (partner-only endpoints such as GET /api/platform/maintenance). Scaled by your rate tier like wallet buckets.

​

Limits

​

Trading (writes)

​

Order reads

​

Market data

​

Portfolio (/me/*)

​

Discovery

​

Platform

​

Counter behaviour

​

Response headers

X-RateLimit-Limit: 100
X-RateLimit-Remaining: 87
X-RateLimit-Reset: 60

• X-RateLimit-Reset is delta-seconds, not a Unix epoch. On fixed-window buckets it is the seconds until the window resets (60 means “the limit resets in 60 seconds”). On the token-bucket order endpoints it is 0 while you still have budget and, on a 429, the seconds until the next request fits.
• The HTTP header Retry-After is set to the same delta-seconds on every 429.
• On 429 Too Many Requests, the response body uses the standard error envelope:

{
  "status": 429,
  "error": {
    "code": "rate_limited",
    "message": "rate limit exceeded for orders-open",
    "trace_id": "...",
    "retryAfterSec": 12
  }
}

​

Retry guidance

1. On 429: wait Retry-After (header value, in seconds) before retrying. Do not retry sooner — the bucket has not refilled.
2. Backoff on 5xx: exponential backoff with jitter, starting at 200ms, capped at 10s. Maximum 3 attempts for idempotent calls (GET, DELETE). Never re-sign a write to retry it after a 5xx — a new signature is a second order or withdrawal. Resend the identical request instead. For POST /api/orders/place and /place-batch, the same signed order under the same clientOrderId returns the original result (see placement_unknown in Error codes). For POST /api/withdrawals/request, the same body with the same signed salt never creates a second withdrawal.
3. Circuit-break on persistent 5xx: if your error rate exceeds 50% over 30 seconds, stop sending new requests for 60 seconds.

​

Higher limits

• wallet-scoped ceilings are raised through your partner rate tier (rateLimitPerMin), which scales every wallet-scoped limit together (see the top of this page).
• ip-scoped limits are the same for every caller behind an IP; they are not raised per partner or per IP.
• Platform-wide ceilings cannot be exceeded; they protect the shared matcher.