release: v13.11.10

Author: Innei

apps/admin/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@mx-admin/admin",
   "private": true,
-  "version": "8.3.25",
+  "version": "8.3.26",
   "scripts": {
     "dev": "NODE_ENV=development vite --mode development --open --host",
     "build": "NODE_ENV=production vite build --mode production",

apps/core/CHANGELOG.md

@@ -1,3 +1,10 @@
+## [13.11.10](https://github.com/mx-space/core/compare/v13.11.9...v13.11.10) (2026-07-02)
+
+
+### Bug Fixes
+
+* **admin:** resolve route imports by absolute path in admin-routes plugin ([91357b9](https://github.com/mx-space/core/commit/91357b947))
+
 ## [13.11.9](https://github.com/mx-space/core/compare/v13.11.8...v13.11.9) (2026-07-02)
 
 

apps/core/RELEASE_NOTES.md

@@ -1,16 +1,19 @@
-# v13.11.9
+# v13.11.10
 
 ## TL;DR
 
-Fixes AI providers on custom OpenAI-compatible endpoints such as Gemini, with new per-provider model list URL and /v1 settings.
+Ships the v13.11.9 AI compat-provider fixes whose release build failed, plus the admin build fix that unblocks it.
+
+> v13.11.9 was tagged but its release pipeline failed during the admin build, so no artifacts were published. This release supersedes it; the changes below cover both versions.
 
 ## Changes
 
+- Admin production build works again: route imports generated by the admin-routes plugin now use absolute paths, fixing a resolution failure introduced by the vite 8.1.2 / rolldown 1.1.3 bump ([91357b9](https://github.com/mx-space/core/commit/91357b947))
 - AI providers pointing at custom OpenAI-compatible endpoints (e.g. Gemini's `/v1beta/openai/`) work again: each provider now has a **Model list URL** field (fetches models straight from your endpoint; empty keeps the built-in registry) and an **Append /v1 to base URL** toggle, and the OpenAI-only `store` field is no longer sent to third-party hosts — previously the source of `400 status code (no body)` failures in summary, insights, translation, and connection tests ([#2761](https://github.com/mx-space/core/issues/2761))
 - Admin: the provider drawer exposes the two new settings; the model dropdown lists models fetched from your endpoint ahead of registry presets; model-fetch toasts now report errors and empty results instead of always claiming success ([4d4565b](https://github.com/mx-space/core/commit/4d4565b1f745dd5edec9a2521611c0975759c4bd))
 - Provider **context window** and **max output tokens** values are now actually persisted — they were silently dropped when saving AI settings ([4d4565b](https://github.com/mx-space/core/commit/4d4565b1f745dd5edec9a2521611c0975759c4bd))
 - Internal: the HTTP client layer moved from axios to ofetch with retry/backoff ([fc49c01](https://github.com/mx-space/core/commit/fc49c0175)), and workspace dependencies were refreshed (pi-ai 0.80, vite 8.1.2, better-auth 1.6.23, nodemailer 9, …) ([06ded90](https://github.com/mx-space/core/commit/06ded90d3))
 
 ---
 
-**Full Changelog**: https://github.com/mx-space/core/compare/v13.11.8...v13.11.9
+**Full Changelog**: https://github.com/mx-space/core/compare/v13.11.8...v13.11.10

apps/core/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mx-space/core",
-  "version": "13.11.9",
+  "version": "13.11.10",
   "type": "module",
   "private": true,
   "engines": {