Built-in image generation returns 404 through OpenCodex gateway

[testerggf]

Summary

Built-in image generation fails with HTTP 404 when Codex is routed through the OpenCodex gateway.

The same user request worked previously, but now a simple request such as “generate a cute cat” fails before an image is produced.

Environment

• OpenCodex.app: 1.1.0
• Gateway health: CodexBridge Engine V2, version 1.1.0
• Codex model: gpt-5.6-luna
• Managed base URL: http://127.0.0.1:25777/v1
• macOS desktop installation

Reproduction

1. Enable/use the OpenCodex-managed Codex provider.
2. In Codex Desktop, invoke the built-in image generation tool with a minimal prompt, for example:

   Generate one adorable, fluffy kitten.
   

3. Observe:

   image generation failed: http 404 Not Found
   {"error":"Endpoint not found"}
   

The same response can be reproduced locally without invoking a real image model:

curl --noproxy '*' -i -X POST \
  http://127.0.0.1:25777/v1/images/generations \
  -H 'Content-Type: application/json' \
  --data '{}'

Response:

HTTP/1.1 404 Not Found

{"error":"Endpoint not found"}

Evidence

• The active Codex configuration routes requests through the local gateway at http://127.0.0.1:25777/v1.
• The installed gateway returns {"error":"Endpoint not found"} for unmatched routes.
• The installed bundle contains a native image bridge, but it sends image requests to https://chatgpt.com/backend-api/codex/responses.
• The bridge is only reached from the internal /v1/responses flow after the provider emits the internal opencodex_generate_image tool call.
• There is no /v1/images/generations route in the gateway.

Relevant installed bundle locations:

• Contents/Resources/dist/server/gateway.js: fallback 404 route
• Contents/Resources/dist/server/router.js: internal image-tool handling
• Contents/Resources/dist/services/native_image_bridge.js: native image bridge

Expected behavior

Built-in image generation should either:

1. be forwarded through the existing native /v1/responses image bridge, or
2. expose a compatible POST /v1/images/generations route that translates the request to the native bridge.

Actual behavior

The built-in image-generation request reaches the local OpenCodex gateway, hits an unimplemented image endpoint, and fails with 404 before any image is generated.

This looks like a routing/compatibility regression introduced by the managed gateway path rather than a prompt, model quota, or API-key problem.

[AITabby]

Fixed in OpenCodex v1.1.2.

The managed gateway now supports POST /v1/images/generations and forwards the native image request through the Codex Images API instead of returning the local 404.

The native image path uses gpt-image-2, matching the official Codex image-generation implementation. Third-party Responses models use the same native image executor; gpt-image-1.5 is retained only as an explicit fallback when image2 is unavailable.

Verified:

• Official/native image generation succeeds through the gateway.
• A third-party model successfully completed an image_generation_call through the gateway.
• Full test suite: 167/167 passed.
• Release: https://github.com/AITabby/opencodex/releases/tag/v1.1.2