# OpenAI-Compatible Codex Gateway

Codex Pooler can act as a narrow OpenAI-compatible Codex gateway for selected SDK clients. Clients call `/v1` with a Pool API key, and Codex Pooler translates supported requests into Codex-compatible work before applying the same Pool routing, limits, account selection, and accounting path.

It does not provide full OpenAI API parity.

## Supported Shape At A Glance

<table>
  <thead>
    <tr>
      <th>Client need</th>
      <th>Use this route</th>
      <th>Boundary</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>List models visible to the Pool</td>
      <td><code>GET /v1/models</code></td>
      <td>OpenAI-shaped model list from Pool and catalog state</td>
    </tr>
    <tr>
      <td>Send Responses work</td>
      <td><code>POST /v1/responses</code></td>
      <td>Translated to Codex-compatible work and routed through Pool policy</td>
    </tr>
    <tr>
      <td>Use chat-completions-style clients</td>
      <td><code>POST /v1/chat/completions</code></td>
      <td>Coerced to Codex Responses work and normalized back to chat shape</td>
    </tr>
    <tr>
      <td>Use narrow Responses websocket compatibility</td>
      <td><code>GET /v1/responses</code></td>
      <td>Not OpenAI Realtime SDK support</td>
    </tr>
    <tr>
      <td>Read operator metadata</td>
      <td><code>/mcp</code></td>
      <td>Separate endpoint with operator MCP tokens, not Pool API keys</td>
    </tr>
  </tbody>
</table>

## Base URL

Use `https://codex-pooler.example.com/v1` for deployed examples, or `http://localhost:4000/v1` for local setup. The credential is a Pool API key.

```text
Base URL: https://codex-pooler.example.com/v1
Authorization: Bearer <pool-api-key>
```

## Unsupported Boundaries

Codex Pooler does not support `/v1/realtime` or OpenAI Realtime SDK websocket or session routes. `GET /v1/responses` is narrow Responses websocket compatibility only.

Unsupported `/v1` routes may return deterministic OpenAI-shaped unsupported endpoint errors when explicitly routed. Examples include embeddings, batches, moderations, fine-tuning jobs, image variations, response retrieve, response cancel, response delete, and `/v1/responses/compact`.

## Canonical Setup Pages

- [OpenAI-Compatible SDKs](/clients/openai-compatible/) explains SDK base URLs, route support, and MCP separation
- [Runtime Routes](/reference/runtime-routes/) lists supported, translated, partial, and unsupported public routes
- [Routing Strategies](/reference/routing-strategies/) explains how `/v1` traffic still uses Pool eligibility and routing

## Related Discovery Pages

- [AI coding agent gateway](/discovery/ai-coding-agent-gateway/)
- [Self-hosted Codex gateway](/discovery/self-hosted-codex-gateway/)
- [Codex account pooling](/discovery/codex-account-pooling/)
- [Codex Pooler vs direct credentials](/discovery/codex-pooler-vs-direct-credentials/)