Terminus-Gitbot · Sep 30, 2026 · Sep 30, 2026
diff --git a/litellm/Chart.yaml b/litellm/Chart.yaml
@@ -1,6 +1,6 @@
 apiVersion: v2
-appVersion: "1.83.3"
+appVersion: "1.103.0"
 description: LiteLLM Proxy - Call 100+ LLMs in OpenAI format
 name: litellm
 type: application
-version: 1.0.21
+version: 1.0.22
diff --git a/litellm/OlaresManifest.yaml b/litellm/OlaresManifest.yaml
@@ -11,15 +11,15 @@ metadata:
   description: Call 100+ LLMs in OpenAI format
   appid: litellm
   title: LiteLLM
-  version: "1.0.21"
+  version: "1.0.22"
   categories:
   - Utilities_v112
   - models
   tags:
   - gateway
 
 spec:
-  versionName: "1.83.3"
+  versionName: "1.103.0"
   featuredImage: https://app.cdn.olares.com/appstore/litellm/1.webp
   promoteImage:
     - https://app.cdn.olares.com/appstore/litellm/1.webp
@@ -39,41 +39,12 @@ spec:
     - Multi-tenant: Support for multiple users and teams
 
   upgradeDescription: |
+    Chart 1.0.22 upgrades LiteLLM Proxy from v1.83.3-stable to v1.103.0.
+    Image: docker.io/beclab/berriai-litellm:v1.103.0. The same digest is published as the main-stable tag.
 
-    Run the official OpenResty entrance as UID/GID 1000 with writable runtime paths.
-    Switch the reverse-proxy image from Bitnami OpenResty (`beclab/aboveos-bitnami-openresty:1.25.3-2`) to the official OpenResty image `openresty/openresty:1.29.2.5-bookworm-fat`.
-
-    Align nginx config mounts and log paths with the official image layout (`/etc/nginx/conf.d/default.conf`, `/usr/local/openresty/nginx/logs/...`), and drop Bitnami-specific `OPENRESTY_CONF_FILE` wiring.
-
-    v1.0.14: Move `DATABASE_URL`, `UI_PASSWORD`, and `SMTP_PASSWORD` into Secret (`secretKeyRef`).
-
-    Upgrade to [v1.83.3-stable](https://github.com/BerriAI/litellm/releases/tag/v1.83.3-stable) (from v1.81.4).
-
-    Security:
-    - Fix privilege escalation on /key/block, /key/unblock, /key/update max_budget
-    - Global secret redaction via root logger and key-name-based pattern matching
-    - Supply chain hardening: pin all Docker/npm/pip dependencies, cosign image signing
-
-    Providers:
-    - Anthropic: preserve thinking.summary when routing to OpenAI Responses API; add opt-out flag for default reasoning summary
-    - Gemini: fix image token undercounting in usage metadata; pass model to context caching URL builder for custom api_base
-    - Azure: auto-route gpt-5.4+ tools+reasoning to Responses API
-    - Bedrock: support cache_control_injection_points for tool_config; respect api_base in count_tokens
-    - OpenAI: round-trip Responses API reasoning_items in chat completions; GPT-5.4 mini/nano day-0 support
-
-    *Proxy & Core:*
-    - Project-level guardrails support
-    - Redis cluster caching fix; Redis-backed cross-pod budget enforcement
-    - Fix streaming guardrail race condition and post-call guardrail response capture
-    - Per-model-group deployment affinity; per-model rate limits for teams
-    - Auto-recover shared aiohttp session when closed
-
-    UI:
-    - Migrate components from Tremor to Ant Design
-    - Team management modernization; team guardrails UI
-    - Logs filter fix (empty filter no longer shows stale data); CSV export fix on Global Usage page
-    - Paginated team search on usage page; key alias substring search
+    The master key, admin UI credentials, PostgreSQL wiring, and proxy config stay as they are. LiteLLM applies Prisma migrations on startup. When public._prisma_migrations already exists, an init container writes a PostgreSQL custom-format dump to appData/backups/chart-1.0.22/litellm.dump first. Later starts and a database without that table skip the dump.
 
+    Release notes: https://github.com/BerriAI/litellm/releases/tag/v1.103.0
   developer: BerriAI
   website: https://www.litellm.ai
   sourceCode: https://github.com/BerriAI/litellm

diff --git a/litellm/i18n/de-DE/OlaresManifest.yaml b/litellm/i18n/de-DE/OlaresManifest.yaml
@@ -15,33 +15,7 @@ spec:
     - Proxy Server: Als AI Gateway/Proxy-Server bereitstellen
     - Multi-tenant: Unterstützung für mehrere Benutzer und Teams
   upgradeDescription: |
-    Umstellung des Reverse-Proxy-Images von Bitnami OpenResty (`beclab/aboveos-bitnami-openresty:1.25.3-2`) auf das offizielle OpenResty-Image `openresty/openresty:1.29.2.5-bookworm-fat`.
-
-    Anpassung der nginx-Konfigurations-Mounts und Log-Pfade an das Layout des offiziellen Images (`/etc/nginx/conf.d/default.conf`, `/usr/local/openresty/nginx/logs/...`) und Entfernung der Bitnami-spezifischen `OPENRESTY_CONF_FILE`-Verdrahtung.
-
-    Upgrade auf [v1.83.3-stable](https://github.com/BerriAI/litellm/releases/tag/v1.83.3-stable) (von v1.81.4).
-
-    Sicherheit:
-    - Privilege Escalation auf /key/block, /key/unblock, /key/update max_budget behoben
-    - Globale Secret-Redaktion über Root-Logger und schlüsselnamensbasierte Mustermatching
-    - Supply-Chain-Härtung: alle Docker-/npm-/pip-Abhängigkeiten pinnen, Cosign-Image-Signing
-
-    Anbieter:
-    - Anthropic: thinking.summary beim Routing zur OpenAI Responses API beibehalten; Opt-out-Flag für Standard-Reasoning-Summary
-    - Gemini: Unterzählung von Image-Tokens in Usage-Metadaten behoben; Modell an Context-Caching-URL-Builder für custom api_base übergeben
-    - Azure: gpt-5.4+ tools+reasoning automatisch zur Responses API routen
-    - Bedrock: cache_control_injection_points für tool_config; api_base in count_tokens respektieren
-    - OpenAI: Responses-API-reasoning_items in Chat Completions round-trippen; GPT-5.4 mini/nano Day-0-Support
-
-    *Proxy & Core:*
-    - Guardrails auf Projektebene
-    - Redis-Cluster-Caching-Fix; Redis-basiertes Cross-Pod-Budget-Enforcement
-    - Streaming-Guardrail-Race Condition und Post-Call-Guardrail-Response-Capture behoben
-    - Deployment-Affinität pro Modellgruppe; Rate Limits pro Modell für Teams
-    - Auto-Recovery der shared aiohttp-Session bei Schließung
-
-    UI:
-    - Komponenten von Tremor nach Ant Design migriert
-    - Team-Management modernisiert; Team-Guardrails-UI
-    - Logs-Filter-Fix (leerer Filter zeigt keine veralteten Daten mehr); CSV-Export-Fix auf Global Usage
-    - Paginierte Teamsuche auf Usage-Seite; Key-Alias-Teilstringsuche
+    Chart 1.0.22 aktualisiert den LiteLLM Proxy von v1.83.3-stable auf v1.103.0.
+    Image: docker.io/beclab/berriai-litellm:v1.103.0. Derselbe Digest ist als main-stable veroeffentlicht.
+    Master Key, Admin-Zugang, PostgreSQL-Anbindung und Proxy-Konfiguration bleiben unveraendert. LiteLLM fuehrt Prisma-Migrationen beim Start aus. Existiert public._prisma_migrations, schreibt ein Init-Container vorher ein pg_dump-Archiv nach appData/backups/chart-1.0.22/litellm.dump. Spaetere Starts und eine Datenbank ohne diese Tabelle ueberspringen das.
+    Release Notes: https://github.com/BerriAI/litellm/releases/tag/v1.103.0
diff --git a/litellm/i18n/en-US/OlaresManifest.yaml b/litellm/i18n/en-US/OlaresManifest.yaml
@@ -16,33 +16,9 @@ spec:
     - Multi-tenant: Support for multiple users and teams
 
   upgradeDescription: |
-    Switch the reverse-proxy image from Bitnami OpenResty (`beclab/aboveos-bitnami-openresty:1.25.3-2`) to the official OpenResty image `openresty/openresty:1.29.2.5-bookworm-fat`.
+    Chart 1.0.22 upgrades LiteLLM Proxy from v1.83.3-stable to v1.103.0.
+    Image: docker.io/beclab/berriai-litellm:v1.103.0. The same digest is published as the main-stable tag.
 
-    Align nginx config mounts and log paths with the official image layout (`/etc/nginx/conf.d/default.conf`, `/usr/local/openresty/nginx/logs/...`), and drop Bitnami-specific `OPENRESTY_CONF_FILE` wiring.
+    The master key, admin UI credentials, PostgreSQL wiring, and proxy config stay as they are. LiteLLM applies Prisma migrations on startup. When public._prisma_migrations already exists, an init container writes a PostgreSQL custom-format dump to appData/backups/chart-1.0.22/litellm.dump first. Later starts and a database without that table skip the dump.
 
-    Upgrade to [v1.83.3-stable](https://github.com/BerriAI/litellm/releases/tag/v1.83.3-stable) (from v1.81.4).
-
-    Security:
-    - Fix privilege escalation on /key/block, /key/unblock, /key/update max_budget
-    - Global secret redaction via root logger and key-name-based pattern matching
-    - Supply chain hardening: pin all Docker/npm/pip dependencies, cosign image signing
-
-    Providers:
-    - Anthropic: preserve thinking.summary when routing to OpenAI Responses API; add opt-out flag for default reasoning summary
-    - Gemini: fix image token undercounting in usage metadata; pass model to context caching URL builder for custom api_base
-    - Azure: auto-route gpt-5.4+ tools+reasoning to Responses API
-    - Bedrock: support cache_control_injection_points for tool_config; respect api_base in count_tokens
-    - OpenAI: round-trip Responses API reasoning_items in chat completions; GPT-5.4 mini/nano day-0 support
-
-    *Proxy & Core:*
-    - Project-level guardrails support
-    - Redis cluster caching fix; Redis-backed cross-pod budget enforcement
-    - Fix streaming guardrail race condition and post-call guardrail response capture
-    - Per-model-group deployment affinity; per-model rate limits for teams
-    - Auto-recover shared aiohttp session when closed
-
-    UI:
-    - Migrate components from Tremor to Ant Design
-    - Team management modernization; team guardrails UI
-    - Logs filter fix (empty filter no longer shows stale data); CSV export fix on Global Usage page
-    - Paginated team search on usage page; key alias substring search
+    Release notes: https://github.com/BerriAI/litellm/releases/tag/v1.103.0
diff --git a/litellm/i18n/es-ES/OlaresManifest.yaml b/litellm/i18n/es-ES/OlaresManifest.yaml
@@ -15,33 +15,7 @@ spec:
     - Proxy Server: Despliega como AI Gateway/Proxy
     - Multi-tenant: Soporte para varios usuarios y equipos
   upgradeDescription: |
-    Cambio de la imagen del reverse proxy de Bitnami OpenResty (`beclab/aboveos-bitnami-openresty:1.25.3-2`) a la imagen oficial de OpenResty `openresty/openresty:1.29.2.5-bookworm-fat`.
-
-    Alineación de los montajes de configuración de nginx y las rutas de registros con el diseño de la imagen oficial (`/etc/nginx/conf.d/default.conf`, `/usr/local/openresty/nginx/logs/...`), y eliminación del cableado específico de Bitnami `OPENRESTY_CONF_FILE`.
-
-    Actualización a [v1.83.3-stable](https://github.com/BerriAI/litellm/releases/tag/v1.83.3-stable) (desde v1.81.4).
-
-    Seguridad:
-    - Corrección de escalada de privilegios en /key/block, /key/unblock, /key/update max_budget
-    - Redacción global de secretos vía root logger y coincidencia por nombre de clave
-    - Endurecimiento de la cadena de suministro: fijar todas las dependencias Docker/npm/pip, firma de imágenes cosign
-
-    Proveedores:
-    - Anthropic: conservar thinking.summary al enrutar a OpenAI Responses API; flag de exclusión para el resumen de razonamiento por defecto
-    - Gemini: corregir el infraconteo de tokens de imagen en metadatos de uso; pasar el modelo al constructor de URL de context caching para api_base personalizado
-    - Azure: autoenrutar gpt-5.4+ tools+reasoning a Responses API
-    - Bedrock: soportar cache_control_injection_points para tool_config; respetar api_base en count_tokens
-    - OpenAI: round-trip de reasoning_items de Responses API en chat completions; soporte day-0 de GPT-5.4 mini/nano
-
-    *Proxy & Core:*
-    - Guardrails a nivel de proyecto
-    - Corrección de caché Redis cluster; aplicación de presupuesto entre pods respaldada por Redis
-    - Corrección de condición de carrera de guardrail en streaming y captura de respuesta post-call
-    - Afinidad de despliegue por grupo de modelos; límites de tasa por modelo para equipos
-    - Autorrecuperación de la sesión aiohttp compartida cuando se cierra
-
-    UI:
-    - Migración de componentes de Tremor a Ant Design
-    - Modernización de gestión de equipos; UI de guardrails de equipo
-    - Corrección del filtro de logs (filtro vacío ya no muestra datos obsoletos); corrección de exportación CSV en Global Usage
-    - Búsqueda paginada de equipos en usage; búsqueda por subcadena de alias de clave
+    El chart 1.0.22 actualiza LiteLLM Proxy de v1.83.3-stable a v1.103.0.
+    Imagen: docker.io/beclab/berriai-litellm:v1.103.0. El mismo digest esta publicado como main-stable.
+    La master key, las credenciales de la UI, PostgreSQL y la configuracion del proxy no cambian. LiteLLM aplica las migraciones de Prisma al arrancar. Si ya existe public._prisma_migrations, un init container escribe antes un volcado pg_dump en appData/backups/chart-1.0.22/litellm.dump. Los arranques posteriores y una base sin esa tabla lo omiten.
+    Notas: https://github.com/BerriAI/litellm/releases/tag/v1.103.0
diff --git a/litellm/i18n/fr-FR/OlaresManifest.yaml b/litellm/i18n/fr-FR/OlaresManifest.yaml
@@ -15,33 +15,7 @@ spec:
     - Proxy Server : Déployez comme AI Gateway/Proxy
     - Multi-tenant : Prise en charge de plusieurs utilisateurs et équipes
   upgradeDescription: |
-    Remplacement de l'image du reverse proxy Bitnami OpenResty (`beclab/aboveos-bitnami-openresty:1.25.3-2`) par l'image officielle OpenResty `openresty/openresty:1.29.2.5-bookworm-fat`.
-
-    Alignement des montages de configuration nginx et des chemins de journaux sur la disposition de l'image officielle (`/etc/nginx/conf.d/default.conf`, `/usr/local/openresty/nginx/logs/...`), et suppression du câblage spécifique Bitnami `OPENRESTY_CONF_FILE`.
-
-    Mise à niveau vers [v1.83.3-stable](https://github.com/BerriAI/litellm/releases/tag/v1.83.3-stable) (depuis v1.81.4).
-
-    Sécurité :
-    - Correction d'une escalade de privilèges sur /key/block, /key/unblock, /key/update max_budget
-    - Rédaction globale des secrets via le root logger et correspondance par nom de clé
-    - Durcissement de la chaîne d'approvisionnement : épinglage de toutes les dépendances Docker/npm/pip, signature d'images cosign
-
-    Fournisseurs :
-    - Anthropic : conserver thinking.summary lors du routage vers OpenAI Responses API ; flag d'opt-out pour le résumé de raisonnement par défaut
-    - Gemini : corriger le sous-comptage des tokens image dans les métadonnées d'usage ; passer le modèle au constructeur d'URL de context caching pour api_base personnalisé
-    - Azure : routage auto de gpt-5.4+ tools+reasoning vers Responses API
-    - Bedrock : prise en charge de cache_control_injection_points pour tool_config ; respecter api_base dans count_tokens
-    - OpenAI : round-trip des reasoning_items Responses API dans les chat completions ; support day-0 GPT-5.4 mini/nano
-
-    *Proxy & Core :*
-    - Guardrails au niveau projet
-    - Correctif de cache Redis cluster ; application budgétaire cross-pod basée sur Redis
-    - Correctif de condition de course des guardrails en streaming et capture de réponse post-call
-    - Affinité de déploiement par groupe de modèles ; limites de débit par modèle pour les équipes
-    - Récupération auto de la session aiohttp partagée lorsqu'elle est fermée
-
-    UI :
-    - Migration des composants de Tremor vers Ant Design
-    - Modernisation de la gestion d'équipes ; UI des guardrails d'équipe
-    - Correctif du filtre de logs (filtre vide n'affiche plus de données obsolètes) ; correctif d'export CSV sur Global Usage
-    - Recherche d'équipes paginée sur la page usage ; recherche par sous-chaîne d'alias de clé
+    Le chart 1.0.22 met a jour LiteLLM Proxy de v1.83.3-stable vers v1.103.0.
+    Image : docker.io/beclab/berriai-litellm:v1.103.0. Le meme digest est publie comme main-stable.
+    La master key, les identifiants de l interface, PostgreSQL et la configuration du proxy restent identiques. LiteLLM applique les migrations Prisma au demarrage. Si public._prisma_migrations existe deja, un init container ecrit d abord une archive pg_dump dans appData/backups/chart-1.0.22/litellm.dump. Les demarrages suivants et une base sans cette table ignorent cette copie.
+    Notes : https://github.com/BerriAI/litellm/releases/tag/v1.103.0
diff --git a/litellm/i18n/it-IT/OlaresManifest.yaml b/litellm/i18n/it-IT/OlaresManifest.yaml
@@ -15,33 +15,7 @@ spec:
     - Proxy Server: Distribuisci come AI Gateway/Proxy
     - Multi-tenant: Supporto per più utenti e team
   upgradeDescription: |
-    Sostituzione dell'immagine del reverse proxy da Bitnami OpenResty (`beclab/aboveos-bitnami-openresty:1.25.3-2`) all'immagine ufficiale OpenResty `openresty/openresty:1.29.2.5-bookworm-fat`.
-
-    Allineamento dei mount della configurazione nginx e dei percorsi dei log al layout dell'immagine ufficiale (`/etc/nginx/conf.d/default.conf`, `/usr/local/openresty/nginx/logs/...`) e rimozione del cablaggio specifico Bitnami `OPENRESTY_CONF_FILE`.
-
-    Aggiornamento a [v1.83.3-stable](https://github.com/BerriAI/litellm/releases/tag/v1.83.3-stable) (da v1.81.4).
-
-    Sicurezza:
-    - Fix escalation di privilegi su /key/block, /key/unblock, /key/update max_budget
-    - Redazione globale dei secret via root logger e pattern matching basato sul nome chiave
-    - Hardening della supply chain: pin di tutte le dipendenze Docker/npm/pip, firma immagini cosign
-
-    Provider:
-    - Anthropic: preservare thinking.summary quando si instrada a OpenAI Responses API; flag di opt-out per il summary di reasoning predefinito
-    - Gemini: fix sottoconteggio token immagine nei metadati usage; passare il modello al builder URL di context caching per api_base personalizzato
-    - Azure: auto-routing di gpt-5.4+ tools+reasoning a Responses API
-    - Bedrock: supporto cache_control_injection_points per tool_config; rispettare api_base in count_tokens
-    - OpenAI: round-trip di reasoning_items Responses API in chat completions; supporto day-0 GPT-5.4 mini/nano
-
-    *Proxy & Core:*
-    - Supporto guardrail a livello di progetto
-    - Fix caching Redis cluster; enforcement budget cross-pod basato su Redis
-    - Fix race condition guardrail in streaming e cattura risposta post-call
-    - Affinità di deployment per gruppo modelli; rate limit per modello per i team
-    - Auto-recupero della sessione aiohttp condivisa quando chiusa
-
-    UI:
-    - Migrazione componenti da Tremor ad Ant Design
-    - Modernizzazione gestione team; UI guardrail di team
-    - Fix filtro log (filtro vuoto non mostra più dati obsoleti); fix export CSV su Global Usage
-    - Ricerca team paginata nella pagina usage; ricerca per sottostringa alias chiave
+    Il chart 1.0.22 aggiorna LiteLLM Proxy da v1.83.3-stable a v1.103.0.
+    Immagine: docker.io/beclab/berriai-litellm:v1.103.0. Lo stesso digest e pubblicato come main-stable.
+    La master key, le credenziali dell interfaccia, PostgreSQL e la configurazione del proxy restano invariati. LiteLLM applica le migrazioni Prisma all avvio. Se public._prisma_migrations esiste gia, un init container scrive prima un dump pg_dump in appData/backups/chart-1.0.22/litellm.dump. Gli avvii successivi e un database senza quella tabella saltano la copia.
+    Note: https://github.com/BerriAI/litellm/releases/tag/v1.103.0