Network Access
Network Access limits which IP addresses can call the Gateway. Requests from anywhere else are rejected before they reach a model or a tool, even when they carry a valid application API key.
Required role
Applies to
Why use it
Use Network Access when applications should accept Gateway requests only from known networks, such as your corporate VPN or production servers.
- A second factor for every request. A caller needs both a valid key and an allowed network.
- Enforced centrally. The rule lives in Optiak, not in each application, and no application can bypass it.
- Scoped to what needs it. Lock down production applications without blocking teams experimenting elsewhere.
How it works
After the Gateway identifies the application from its API key, it checks the address the request comes from against every allowlist that covers that application. This happens before guardrails, model choice or any tool call, for every endpoint of the OpenAI-compatible API and for the MCP endpoint.
If at least one allowlist covers the application but the address matches none of them, the Gateway rejects the request with 403 Forbidden and the message Request originates from an IP address outside the organization's allowed range. No model or tool is called.
IP allowlists
An allowlist is a named list of IPv4 or IPv6 addresses and CIDR ranges. Include the public IP addresses the Gateway sees when your applications connect, such as your office or VPN's outbound address, your NAT gateway's public address or your servers' public IPs. Give each allowlist a name your team recognizes, such as "Corporate VPN" or "Production cluster", and a description of where it applies.
Scope
Each allowlist applies to one of these:
- The whole organization: every application, including those created later.
- Selected projects and applications: choose projects, individual applications or both. The allowlist covers every selected application and every application in the selected projects, including applications created in those projects later.
An application that no allowlist covers accepts requests from any address.
Several allowlists
When more than one allowlist covers an application, a request is allowed if its address matches any of them. Each allowlist adds trusted networks; none of them narrows another. For example, an organization-wide "Corporate VPN" allowlist and a "CI runners" allowlist on one project let that project's applications accept traffic from both.
Set it up
In the console, open Company setup, then Security, and select Create allowlist. Give it a name, enter one address or range per line under Allowed IP ranges and choose its Scope.
The console itself isn't affected by allowlists, so you can always edit or delete one that blocks traffic it shouldn't.
Limits and behavior
- Ranges use the network address.
203.0.113.0/24is valid;203.0.113.5/24is rejected when you save. A range that covers every address, such as0.0.0.0/0, and duplicate entries are rejected too. - Write IPv4 addresses in plain form (
10.0.0.0/8), not as IPv4-mapped IPv6 ranges. - Deleting an allowlist stops enforcing it immediately. If it was the only one covering an application, that application accepts requests from any address again.
- Every change is recorded in Audit Logs, with who made it and when.
Related
- Data Governance: control where requests are processed.
- Authentication: create and revoke application API keys.
- Organization and Projects: how projects group applications.
- Using the API: endpoints and error responses.