Which AI applications can go through the AI Gateway
Application by application, whether its requests can pass through the WASViking AI Gateway today, why some sign-in modes cannot be routed by any gateway, and which AI Guardian control covers each case instead.
The WASViking® AI Gateway stands between an application and the model provider it calls. It can only do that where the application lets an administrator set the address it calls. This page lists the common applications, says whether the gateway applies today, and names the AI Guardian control that covers the ones it does not.
Use it before you promise a team that "everything will go through the gateway". For most teams the honest answer is "the applications on the company key, plus Claude Code on seats", with the browser extension and the device inventory covering the rest.
How applications reach a model
There are two ways an application reaches a model, and they decide everything on this page.
- With the company's own provider key. The application has a setting for the provider address. Point it at the gateway and every request passes through it, with your policies, your budgets and the request log. Billing is per token on your key.
- With the vendor's sign-in (seats). The person signs in to the vendor's own service, and the application talks to that service directly. There is no address to change, so no gateway, from any supplier, can stand in the middle. Some of these applications also pin their connection to the vendor's servers, so a network proxy cannot read it either. This is how the vendors designed these products, and it is the same for every customer.
The one documented exception is Claude Code on Claude Team or Enterprise seats. Its vendor supports a gateway path for that mode, and the WASViking gateway implements it: see Protect Claude Code on Claude seats.
A Claude Enterprise organization has a second path that is not a gateway at all: its inference hooks send every request, from every Claude application including the desktop app, to WASViking before the model runs. See Protect Claude Enterprise with inference hooks. Claude Team has no inference hooks; compare what each Claude plan allows.
The table
| Application | How people sign in | Through the AI Gateway | Notes, and what covers it instead |
|---|---|---|---|
| Any application or script using the provider SDK | Company API key | Yes | OpenAI, Anthropic, Gemini, Azure OpenAI and OpenAI-compatible endpoints. See Protect model API calls. |
| Claude Code (terminal, VS Code) | Company Anthropic API key | Yes | See Route Claude Code through the AI Gateway. |
| Claude Code (terminal, VS Code) | Claude Team or Enterprise seat | Yes | The vendor documents this gateway path. See Protect Claude Code on Claude seats. |
| Claude Desktop (Chat, Cowork, Code) | Claude Enterprise organization account | Not the gateway: yes through Claude Enterprise Protection | The application has no gateway setting, but the organization's inference hooks send every request to WASViking before the model runs. See Protect Claude Enterprise with inference hooks. |
| Claude Desktop (Chat, Cowork, Code) | Personal Claude account, or a Team organization | No | No gateway setting and no inference hooks; by the vendor's design the session goes straight to its service. Covered by the AI Guardian browser extension on claude.ai and by Desktop AI apps: inventory, the organization-only sign-in package and app control. Compare the two plans in Claude Team or Claude Enterprise. |
| Claude Desktop in third-party inference mode | Company API key, or the company's own single sign-on | Supported with a manual profile; a guide is in preparation | The vendor's mode for organizations that route inference themselves. Conversations stay on the device and billing is per token. The gateway's Anthropic endpoint serves it; list the allowed models in the profile. |
| Codex CLI, and Codex inside ChatGPT Desktop | Company OpenAI API key | Yes | Set the gateway as the model provider in the Codex configuration file. |
| Codex CLI, and Codex inside ChatGPT Desktop | ChatGPT account or workspace seat | No | The sign-in session goes to the vendor's service; no provider address applies. |
| ChatGPT Desktop (chat) | ChatGPT account or workspace seat | No | No gateway setting, and the application pins its connection, so network proxies cannot inspect it either. The vendor's allowed-workspace header keeps personal accounts out but does not route traffic. Covered by the AI Guardian browser extension on chatgpt.com and by Desktop AI apps. |
| IDE assistants that accept a custom endpoint | Company API key | Yes | Any assistant with a provider address and key setting. |
| IDE assistants on the vendor's own subscription | Vendor account | No | Covered by Assistant Guard where the assistant supports managed hooks, and by Desktop AI apps. |
| Gemini CLI and other Google developer tools | Company Gemini API key | Yes | The Gemini endpoint of the gateway. |
| Gemini CLI and other Google developer tools | Google account | No | Same reason as the other sign-in modes. |
| ChatGPT, Claude, Gemini and other assistants in the browser | Any | Not a gateway case | Covered by the AI Guardian browser extension: the same policies, masking of secrets and record, per person. See Deploy AI Guardian. |
"Yes" means the gateway receives every model request of that application, so your policies, budgets, rate limits and the request log apply to all of it.
What to tell your teams
- Developers on Claude seats keep their seat and go through the gateway with Claude Code. Nothing changes in how they sign in.
- Desktop applications signed in with an account are governed by the browser extension and by Desktop AI apps, not by the gateway. If a team needs the gateway on a desktop application, the vendor's API-key or third-party mode is the path, and billing moves from seats to tokens.
- Applications and scripts on the company key go through the gateway, whatever the vendor. They change the base URL and the key, and nothing else.
Vendor references
- Claude Code: the vendor's page on using an LLM gateway, at
code.claude.com/docs/en/llm-gateway. - Claude Desktop: the vendor's page on the third-party inference mode with an
LLM gateway, at
claude.com/docs/third-party/claude-desktop/gateway.
What next
- Your first protected request through the AI Gateway
- Route Claude Code through the AI Gateway
- Protect Claude Code on Claude seats
- Deploy AI Guardian for the browser extension and Desktop AI apps
- Protect coding assistants with Assistant Guard
- Protect Claude Enterprise with inference hooks