Get a header rule

Content type: application/json

• idinteger
• domain_idinteger
• record_idinteger

  Deprecated single-record scope. Prefer record_ids. Absent for zone-wide rules.

• record_idsarray of integer

  The proxied DNS records this rule applies to. Empty or absent means zone-wide — every proxied record of the domain.

• typestring
  One of: "cache" "drop" "redirect" "rewrite" "waf" "captcha" "rate_limit" "bot_route" "origin_pool" "origin_route" "fingerprint" "error_page" "optimize" "basic_auth" "header"
• enabledboolean
• priorityinteger

  Evaluation order; lower runs first. Defaults to 100.

• host_patternstring

  Legacy single-hostname filter, kept for rules written before host_includes existed. It is evaluated as one more entry of host_includes; prefer the lists.

• host_match_typestring

  How the host filter — host_pattern, host_includes and host_excludes — is matched. One strategy covers all three, exactly as one path_match_type covers both path lists.

  The empty string means "no host filter", and is the only valid value when the pattern and both lists are empty. Set a list without a match type and the API defaults it to wildcard.

  A wildcard entry matches subdomains, not the label itself: *.example.com covers shop.example.com but not example.com — the same reading as the DNS wildcard. Add the bare name as its own entry to include it.

  One of: "" "exact" "wildcard" "regex"
• host_includesarray of string

  Hostnames the rule applies to. Empty or absent means every host the scoped record(s) serve — which on a wildcard-proxied zone (*.example.com) is every subdomain.

• host_excludesarray of string

  Hostnames carved back out of host_includes. An exclude always wins over an include, so "everything except staging" is an empty include list plus one exclude.

• ip_includesarray of string

  Client addresses the rule applies to. Empty or absent means every address — the rule is not scoped by IP at all unless ip_excludes is set.

  Entries are returned canonicalised, which may differ from what was sent: CIDR host bits are masked off (10.0.0.5/8 reads back as 10.0.0.0/8), range endpoints are ordered low-to-high, IPv4-mapped IPv6 is unmapped, and duplicates are dropped. The order of the remaining entries is preserved.

  The address matched is the TCP peer seen by the edge. X-Forwarded-For is never consulted, so a visitor cannot spoof a header to put themselves inside or outside a scope.

• ip_excludesarray of string

  Client addresses carved back out of ip_includes. An exclude always wins over an include. Canonicalised on the way in, exactly like ip_includes.

• action_modestring
  • enforce — the rule acts (block, redirect, challenge, …).
  • dry_run — the rule matches and is logged as "would have acted", but the request reaches the origin unchanged. Use it to test a rule safely.

  Not every rule type honours this; cache ignores it.

  One of: "enforce" "dry_run"
• created_atstring (date-time)
• updated_atstring (date-time)
• path_match_typestring

  How path_includes and path_excludes are interpreted.

  One of: "wildcard" "regex"
• path_includesarray of string

  Paths the rule applies to. Defaults to ["/*"] — everything.

• path_excludesarray of string

  Paths carved back out of path_includes.

• namestring

  Optional label for this rule, e.g. "Security headers", shown in the panel's rule table. Purely for identification. Omit or send an empty string for none.

• opsarray of HeaderOp

  The ordered edit list, applied in array order.

  Caps, all validated on save so an oversized rule fails as a clear 400 instead of as a 431 from the origin: 16 ops per rule, 4096 bytes per value, and 8192 bytes summed over name+value of every set/add op — 4 KB leaves room for a real Content-Security-Policy while staying under the 8 KB per-header limit nginx and Apache default to.

  Headers NSIN owns are refused, per (direction, action, name) rather than by name alone. On the request: Host, the framing and hop-by-hop set (Content-Length, Transfer-Encoding, Connection, Upgrade, Keep-Alive, Proxy-Connection, TE, Trailer), the forwarding set (X-Forwarded-*, X-Real-IP, Forwarded, True-Client-IP, CDN-Loop), the conditional and negotiation set (Cache-Control, If-None-Match, If-Modified-Since, If-Match, If-Range, Range, Accept-Encoding), Sec-WebSocket-*, X-Mafar-*, X-Nsin-* and the whole Nsn-* prefix. On the response: Content-Length, the hop-by-hop set, Content-Encoding, Vary, Server, Strict-Transport-Security (managed by the HSTS ramp on the domain's Security page) and again Nsn-*.

  Three request headers are blocked for set/add but allowed for remove: Cookie, Authorization and Nsn-Connecting-IP. Re-adding a credential is what makes a personalised page cacheable in a shared slot; stripping one before it reaches the origin is a legitimate privacy choice, and is how an owner makes a page cacheable that otherwise would not be.

  Each refusal returns a 400 naming the header, the action and the reason.

  Two response names are accepted but dangerous, and the panel makes a human type the header name to confirm before saving one: Cache-Control, which desyncs the browser's view of freshness from the edge's, and Set-Cookie, which on a cacheable page is the shape of a cross-account session leak. The API applies no confirmation step — an integration writing either one should know why.

  Item fields

  • directionstringrequired
    • request — applied on the way to the origin, after NSIN has stamped its own forwarding headers.
    • response — applied on the way to the visitor, after the response is retrieved from cache or origin. Nothing a response op does reaches the stored copy, so editing a rule takes effect on the very next cache hit rather than when the object expires.

    A 101 WebSocket handshake never passes the response chokepoint, so response ops do not apply to an upgrade.

    One of: "request" "response"
  • actionstringrequired
    • set — replace every existing value with value.
    • add — append value, keeping what the client or origin sent.
    • remove — delete the header. The only action that accepts a trailing * wildcard in name.
    One of: "set" "add" "remove"
  • namestringrequired

    The header field name — an RFC 9110 token, matched case-insensitively. A trailing * (X-Debug-*) is a prefix wildcard and is accepted on remove only; the edge re-checks the blocked list against every name a wildcard actually matches, so a wildcard can never strip a header NSIN owns.

    Some names are refused per (direction, action) rather than outright — see ops on the request body for the list and the reasoning.

  • valuestring

    Required for set and add, and must be empty for remove.

{
  "id": 0,
  "domain_id": 0,
  "record_id": 0,
  "record_ids": [
    0
  ],
  "type": "cache",
  "enabled": true,
  "priority": 0,
  "host_pattern": "string",
  "host_match_type": "",
  "host_includes": [
    "string"
  ],
  "host_excludes": [
    "string"
  ],
  "ip_includes": [
    "string"
  ],
  "ip_excludes": [
    "string"
  ],
  "action_mode": "enforce",
  "created_at": "2026-01-01T00:00:00Z",
  "updated_at": "2026-01-01T00:00:00Z",
  "path_match_type": "wildcard",
  "path_includes": [
    "string"
  ],
  "path_excludes": [
    "string"
  ],
  "name": "string",
  "ops": [
    {
      "direction": "request",
      "action": "set",
      "name": "string",
      "value": "string"
    }
  ]
}

Content type: application/json

• errorstringrequired

  Human-readable description of what went wrong.

{
  "error": "invalid API key"
}

Content type: application/json

• errorstringrequired

  Human-readable description of what went wrong.

{
  "error": "read-only API key"
}

Content type: application/json

• errorstringrequired

  Human-readable description of what went wrong.

{
  "error": "rate limit exceeded"
}