Network allowlist

Neens calls your LLM connections from its own servers. To stop a connection from being used to reach internal systems, Neens refuses any connection whose address is private (a 10.x, 172.16–31.x or 192.168.x address, or a host name that resolves to one) unless that host is on your company’s network allowlist. Add your private model gateway or self-hosted model here and every judge, insight, assistant reply and other LLM feature can use it. Your list applies only to your company; no other company can use it.

At a glance

When you need it

You only need the allowlist when an LLM connection points somewhere that is not on the public internet, for example:

• a model gateway inside your network, such as https://llm-gateway.internal.example.com/v1;
• a self-hosted model server (vLLM, Ollama, LM Studio) on a private address;
• an entire private subnet where your model servers live, such as 10.20.0.0/16.

Public endpoints (OpenAI, Anthropic, Google, OpenRouter, a public LiteLLM proxy) never need an entry.

Add a host in Settings

To remove an entry, select Remove next to it. Changes take effect within a minute.

Manage the list with the API

Use an API key or a session token in the Authorization header. Reading works with any credential that can see the agent; changing a list needs an admin user, so a project API key (nk_live_…) can read the lists but gets 403 on PUT. In the examples below, $ADMIN_TOKEN stands for the session token of a signed-in admin.

curl -s https://neens.example.com/api/settings/llm-allowed-hosts \
  -H "Authorization: Bearer nk_live_…"

{
  "projectId": "project_123",
  "selfServiceEnabled": true,
  "maxEntries": 50,
  "platformManaged": [],
  "company": {
    "entries": [
      {"entry": "10.20.0.0/16", "addedAt": "2026-10-01T09:12:00+00:00", "addedByName": "Ava Lindqvist"},
      {"entry": "llm-gateway.internal.example.com", "addedAt": "2026-10-01T09:12:00+00:00", "addedByName": "Ava Lindqvist"}
    ],
    "canEdit": false
  },
  "project": {"entries": [], "canEdit": false}
}

curl -s -X PUT https://neens.example.com/api/settings/llm-allowed-hosts \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"scope": "company", "entries": ["llm-gateway.internal.example.com", "10.20.0.0/16"]}'

curl -s -X PUT https://neens.example.com/api/settings/llm-allowed-hosts \
  -H "Authorization: Bearer $ADMIN_TOKEN" \
  -H "X-Neens-Project-Id: project_123" \
  -H "Content-Type: application/json" \
  -d '{"scope": "project", "entries": ["vllm.stylist.internal.example.com"]}'

PUT replaces the whole list for that scope with entries (send [] to clear it) and returns the same body as GET. scope defaults to company.

{
  "projectId": "project_123",
  "selfServiceEnabled": true,
  "maxEntries": 50,
  "platformManaged": [],
  "company": {
    "entries": [
      {"entry": "llm-gateway.internal.example.com", "addedAt": "2026-10-02T14:03:00+00:00", "addedByName": "Ava Lindqvist"},
      {"entry": "10.20.0.0/16", "addedAt": "2026-10-02T14:03:00+00:00", "addedByName": "Ava Lindqvist"}
    ],
    "canEdit": true
  },
  "project": {"entries": [], "canEdit": true}
}

{
  "detail": {
    "error": "invalid_entries",
    "message": "'*.example.com': wildcards aren't supported. List each host name exactly.",
    "errors": [
      {
        "index": 0,
        "entry": "*.example.com",
        "code": "wildcard_not_allowed",
        "message": "'*.example.com': wildcards aren't supported. List each host name exactly."
      }
    ]
  }
}

{
  "detail": {
    "error": "company_admin_required",
    "message": "Only a company-wide admin can change the company-wide allowlist."
  }
}

Entry rules

What is always refused

Some addresses are refused even if you list them, because reaching them from Neens’ servers would expose the platform itself rather than your model:

• Cloud metadata endpoints, such as 169.254.169.254 and metadata.google.internal, which hand out server credentials;
• link-local addresses (169.254.0.0/16, fe80::/10);
• loopback addresses, the unspecified address, and multicast;
• address ranges reserved by the platform for its own infrastructure.

The check runs when Neens connects, after looking up the host name, not just when you save. A host name on your list that resolves to one of these addresses is still refused, and so is one that resolves to several addresses if any of them is refused.

Troubleshooting

Related

• LLM connections — create and test the connections the allowlist applies to.
• Members & roles — who counts as an admin.
• Audit log — the egress_allowlist.update history.