fix: rebrand models to Zen, configure publishable key architecture

- Remove deepseek-r1-distill-70b and qwen3-32b from model lists
- Add zen3-nano, zen, zen4-coder-flash to Zen model catalog
- Reorder models by size (nano -> mini -> zen -> pro -> frontier)
- Replace OPENAI_API_KEY with HANZO_API_KEY for publishable key flow
- Replace Deepseek/Kimi/Moonshot display labels with Zen in parsers.ts
- Add HANZO_API_KEY env var to .env.example with IAM documentation

.env.example

@@ -109,6 +109,14 @@ NODE_MAX_OLD_SPACE_SIZE=6144
 
 PROXY=
 
+#===============================#
+# Hanzo API (Zen models)        #
+#===============================#
+# Publishable key for unauthenticated users (rate-limited).
+# Authenticated users get their real API key after IAM login (hanzo.id).
+# Endpoint: https://api.hanzo.ai/v1
+HANZO_API_KEY=
+
 #===================================#
 # Known Endpoints - chat.yaml  #
 #===================================#

config.yaml

@@ -3,110 +3,124 @@
 
 model_list:
   # =====================
-  # zen4 series (Hanzo AI)
+  # Zen model family (Hanzo AI)
+  # Publishable key (HANZO_API_KEY) for unauthenticated users.
+  # Authenticated users override with their own key via IAM (hanzo.id).
   # =====================
-  - model_name: zen4
+  - model_name: zen3-nano
     litellm_params:
-      model: openai/zen4
+      model: openai/zen3-nano
       api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
+      api_key: os.environ/HANZO_API_KEY
     model_info:
-      description: "zen4: General purpose AI"
-
-  - model_name: zen4-pro
-    litellm_params:
-      model: openai/zen4-pro
-      api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
-    model_info:
-      description: "zen4-pro: High capability reasoning"
-
-  - model_name: zen4-max
-    litellm_params:
-      model: openai/zen4-max
-      api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
-    model_info:
-      description: "zen4-max: Extended context window"
+      description: "zen3-nano: 4B dense, fast free-tier model"
 
   - model_name: zen4-mini
     litellm_params:
       model: openai/zen4-mini
       api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
+      api_key: os.environ/HANZO_API_KEY
     model_info:
-      description: "zen4-mini: Fast and efficient"
+      description: "zen4-mini: 8B dense, fast and efficient"
+
+  - model_name: zen
+    litellm_params:
+      model: openai/zen
+      api_base: https://api.hanzo.ai/v1
+      api_key: os.environ/HANZO_API_KEY
+    model_info:
+      description: "zen: 32B default general-purpose model"
+
+  - model_name: zen4-pro
+    litellm_params:
+      model: openai/zen4-pro
+      api_base: https://api.hanzo.ai/v1
+      api_key: os.environ/HANZO_API_KEY
+    model_info:
+      description: "zen4-pro: 80B MoE flagship reasoning"
+
+  - model_name: zen4
+    litellm_params:
+      model: openai/zen4
+      api_base: https://api.hanzo.ai/v1
+      api_key: os.environ/HANZO_API_KEY
+    model_info:
+      description: "zen4: 744B MoE frontier model"
+
+  - model_name: zen4-max
+    litellm_params:
+      model: openai/zen4-max
+      api_base: https://api.hanzo.ai/v1
+      api_key: os.environ/HANZO_API_KEY
+    model_info:
+      description: "zen4-max: 1.04T MoE, 256K context"
 
   - model_name: zen4-ultra
     litellm_params:
       model: openai/zen4-ultra
       api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
+      api_key: os.environ/HANZO_API_KEY
     model_info:
-      description: "zen4-ultra: Maximum reasoning capability"
-
-  - model_name: zen4-coder
-    litellm_params:
-      model: openai/zen4-coder
-      api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
-    model_info:
-      description: "zen4-coder: Code specialist"
+      description: "zen4-ultra: 744B MoE + CoT reasoning"
 
   - model_name: zen4-thinking
     litellm_params:
       model: openai/zen4-thinking
       api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
+      api_key: os.environ/HANZO_API_KEY
     model_info:
-      description: "zen4-thinking: Deep reasoning with chain-of-thought"
+      description: "zen4-thinking: 80B MoE with chain-of-thought"
+
+  - model_name: zen4-coder-flash
+    litellm_params:
+      model: openai/zen4-coder-flash
+      api_base: https://api.hanzo.ai/v1
+      api_key: os.environ/HANZO_API_KEY
+    model_info:
+      description: "zen4-coder-flash: 30B MoE, fast code specialist"
+
+  - model_name: zen4-coder
+    litellm_params:
+      model: openai/zen4-coder
+      api_base: https://api.hanzo.ai/v1
+      api_key: os.environ/HANZO_API_KEY
+    model_info:
+      description: "zen4-coder: 480B MoE code specialist"
 
   - model_name: zen4-coder-pro
     litellm_params:
       model: openai/zen4-coder-pro
       api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
+      api_key: os.environ/HANZO_API_KEY
     model_info:
-      description: "zen4-coder-pro: Premium coding model"
+      description: "zen4-coder-pro: 480B dense BF16 coding model"
 
   # =====================
-  # Standard models (via Hanzo Cloud)
+  # Third-party models (via Hanzo Cloud)
   # =====================
   - model_name: gpt-4o
     litellm_params:
       model: openai/gpt-4o
       api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
+      api_key: os.environ/HANZO_API_KEY
 
   - model_name: gpt-4o-mini
     litellm_params:
       model: openai/gpt-4o-mini
       api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
+      api_key: os.environ/HANZO_API_KEY
 
   - model_name: claude-sonnet-4-5
     litellm_params:
       model: openai/claude-sonnet-4-5
       api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
+      api_key: os.environ/HANZO_API_KEY
 
   - model_name: claude-haiku-4-5
     litellm_params:
       model: openai/claude-haiku-4-5
       api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
-
-  - model_name: deepseek-r1-distill-70b
-    litellm_params:
-      model: openai/deepseek-r1-distill-70b
-      api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
-
-  - model_name: qwen3-32b
-    litellm_params:
-      model: openai/qwen3-32b
-      api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
+      api_key: os.environ/HANZO_API_KEY
 
   # =====================
   # Wildcard passthrough
@@ -115,7 +129,7 @@ model_list:
     litellm_params:
       model: "openai/*"
       api_base: https://api.hanzo.ai/v1
-      api_key: os.environ/OPENAI_API_KEY
+      api_key: os.environ/HANZO_API_KEY
 
 litellm_settings:
   drop_params: true

librechat.yaml

@@ -90,27 +90,30 @@ endpoints:
   # All models routed through api.hanzo.ai OpenAI-compatible gateway
   custom:
     - name: 'Hanzo'
-      apiKey: '${OPENAI_API_KEY}'
+      # Publishable key for unauthenticated users (rate-limited).
+      # Authenticated users override with their own key via IAM (hanzo.id).
+      apiKey: '${HANZO_API_KEY}'
       baseURL: 'https://api.hanzo.ai/v1'
       iconURL: 'https://hanzo.ai/logo/icon.svg'
       models:
         default:
-          # zen4 series (default models)
-          - 'zen4'
-          - 'zen4-pro'
-          - 'zen4-max'
+          # Zen model family (Hanzo AI)
+          - 'zen3-nano'
           - 'zen4-mini'
+          - 'zen'
+          - 'zen4-pro'
+          - 'zen4'
+          - 'zen4-max'
           - 'zen4-ultra'
-          - 'zen4-coder'
           - 'zen4-thinking'
+          - 'zen4-coder-flash'
+          - 'zen4-coder'
           - 'zen4-coder-pro'
-          # Standard models (via Hanzo Cloud)
+          # Third-party models (via Hanzo Cloud)
           - 'gpt-4o'
           - 'gpt-4o-mini'
           - 'claude-sonnet-4-5'
           - 'claude-haiku-4-5'
-          - 'deepseek-r1-distill-70b'
-          - 'qwen3-32b'
         fetch: false
       titleConvo: true
       titleModel: 'zen4-mini'

packages/data-provider/src/parsers.ts

@@ -224,12 +224,8 @@ export const getResponseSender = (endpointOption: Partial<t.TEndpointOption>): s
       return extractOmniVersion(model);
     } else if (model && (model.includes('mistral') || model.includes('codestral'))) {
       return 'Mistral';
-    } else if (model && model.includes('deepseek')) {
-      return 'Deepseek';
-    } else if (model && model.includes('kimi')) {
-      return 'Kimi';
-    } else if (model && model.includes('moonshot')) {
-      return 'Moonshot';
+    } else if (model && (model.startsWith('zen') || model.includes('zen'))) {
+      return 'Zen';
     } else if (model && model.includes('gpt-')) {
       const gptVersion = extractGPTVersion(model);
       return gptVersion || 'GPT';
@@ -265,12 +261,8 @@ export const getResponseSender = (endpointOption: Partial<t.TEndpointOption>): s
       return extractOmniVersion(model);
     } else if (model && (model.includes('mistral') || model.includes('codestral'))) {
       return 'Mistral';
-    } else if (model && model.includes('deepseek')) {
-      return 'Deepseek';
-    } else if (model && model.includes('kimi')) {
-      return 'Kimi';
-    } else if (model && model.includes('moonshot')) {
-      return 'Moonshot';
+    } else if (model && (model.startsWith('zen') || model.includes('zen'))) {
+      return 'Zen';
     } else if (model && model.includes('gpt-')) {
       const gptVersion = extractGPTVersion(model);
       return gptVersion || 'GPT';