Auto Titling for xAI (Grok) and Google (Gemini) Models

[Dylan-Natter]

Auto Titling Not Working for xAI (Grok) and Google (Gemini) Models

Issue Summary

Auto titling functionality does not work for xAI (Grok) and Google (Gemini) models, while working correctly for Anthropic and OpenAI models.

Reported by: User feedback from production environment
Confirmed on: LibreChat latest version (ghcr.io/danny-avila/librechat-dev:latest)

Environment

• LibreChat Version: Latest (ghcr.io/danny-avila/librechat-dev:latest)
• Deployment: Docker Compose
• OS: Linux
• Affected Providers: xAI, Google (Gemini)
• Working Providers: Anthropic, OpenAI

Expected Behavior

Auto titling should work consistently across all AI providers when titleConvo: true is configured.

Actual Behavior

• ❌ xAI Models: No auto titles generated
• ❌ Google Models: No auto titles generated
• ✅ Anthropic: Working correctly
• ✅ OpenAI: Working correctly

Steps to Reproduce

1. Configure xAI and Google endpoints in LibreChat
2. Create new conversations with these models
3. Send several messages back and forth
4. Observe that conversation titles are not automatically generated
5. Compare with Anthropic/OpenAI which generate titles correctly

Root Cause Analysis

xAI Models Issue

• Problem: Parameter compatibility issues with newer Grok models
• Details: xAI API doesn't support presencePenalty and frequencyPenalty parameters
• Grok-4 Specific: New reasoning models (Grok-4-0709) have different parameter requirements and are incompatible with title generation
• Impact: Title generation requests fail silently due to unsupported parameters

Google Models Issue

• Problem: Missing auto titling configuration for standard Google endpoints
• Details: Standard LibreChat Google endpoint configuration lacks explicit title generation settings
• Impact: No title generation attempted for Google/Gemini conversations

Solution/Workaround

For xAI Models:

custom:
  - name: "xAI"
    apiKey: "${XAI_API_KEY}"
    baseURL: "https://api.x.ai/v1"
    models:
      default: ["grok-2-1212", "grok-4-0709"]
    titleConvo: true
    titleModel: "grok-2-1212"  # Use stable model for titles
    dropParams: ["presencePenalty", "frequencyPenalty"]  # Key fix

For Google Models:

custom:
  - name: "Google"
    apiKey: "${GOOGLE_KEY}"
    baseURL: "https://generativelanguage.googleapis.com/v1beta"
    models:
      default: ["gemini-2.5-flash", "gemini-2.5-pro"]
    titleConvo: true
    titleModel: "gemini-2.5-flash"

And disable standard Google endpoint:

ENDPOINTS=openAI,assistants,azureOpenAI,gptPlugins,anthropic

Configuration Details

librechat.yaml changes:

• Added custom xAI endpoint with dropParams configuration
• Added custom Google endpoint with explicit title settings
• Used stable models for title generation (avoid grok-4 compatibility issues)

.env changes:

• Removed google from ENDPOINTS to prevent conflicts with custom endpoint

Testing Results

After implementing the workarounds:

• ✅ xAI Auto Titling: Working with grok-2-1212
• ✅ Google Auto Titling: Working with gemini-2.5-flash
• ✅ No Regressions: Anthropic and OpenAI continue working
• ✅ User Experience: Consistent auto titling across all providers

Suggested Improvements for LibreChat Core

1. Automatic Parameter Compatibility: LibreChat should automatically detect and filter unsupported parameters per provider
2. Standard Endpoint Auto Titling: Auto titling should work with standard endpoint configurations, not require custom endpoints
3. Model-Specific Handling: Better detection and handling of model-specific limitations (e.g., reasoning models)
4. Provider Documentation: Clearer documentation on auto titling configuration requirements per provider
5. Graceful Degradation: When title generation fails, provide user feedback rather than silent failure

Temporary Workarounds Implemented

The following workarounds resolve the issues but require custom endpoint configuration:

• ✅ xAI: Custom endpoint with dropParams and stable model selection
• ✅ Google: Custom endpoint with explicit title configuration
• ✅ Model labeling: Added "(Beta)" to Grok-4 to indicate auto-title limitations

Additional Context

Model-Specific Notes:

• Grok-4: Reasoning model with parameter restrictions - not ideal for title generation
• Grok-2-1212: Stable model, works well for background tasks like titling
• Gemini-2.5-Flash: Fast, efficient for title generation
• Gemini-2.5-Pro: More capable but slower for simple title tasks

Impact: This affects user experience as conversations with xAI and Google models lack descriptive titles, making conversation management difficult.

Files Involved

• librechat.yaml - Endpoint configurations
• .env - Environment variables and endpoint selection
• Auto titling logic for custom vs standard endpoints

Priority

Medium - Affects user experience but has working workarounds

---

Labels: bug, auto-titling, xai, google, configuration, endpoints

[danny-avila]

Please look into all the title configurations to ensure this works for you (from titleConvo and immediately below):

• https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/custom_endpoint#titleconvo
• https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/shared_endpoint_settings

Also, make sure you are indeed using the latest version of LC:

• https://www.librechat.ai/docs/local/docker#update-librechat

I don't have any issues with the following config:

endpoints:
  google:
    titleModel: "gemini-2.5-flash"
  custom:
    - name: "xai"
      apiKey: "${XAI_API_KEY}"
      baseURL: "https://api.x.ai/v1"
      models:
        default: ["grok-3-latest"]
        fetch: true
      titleConvo: true
      titleMethod: "completion"
      titleModel: "grok-3-latest"
      modelDisplayLabel: "Grok"

[Dylan-Natter]

✅ Confirmed Working Solution

Thanks @danny-avila for the quick response and working configuration!

Update:

I can confirm that your recommended solution is working perfectly. After implementing the suggested configuration:

endpoints:
  custom:
    - name: "xAI"
      apiKey: "${XAI_API_KEY}"
      baseURL: "https://api.x.ai/v1"
      models:
        default: ["grok-3-latest"]
        fetch: false
      titleConvo: true
      titleMethod: "completion"
      titleModel: "grok-3-latest"
      modelDisplayLabel: "xAI"
      dropParams: ["presencePenalty", "frequencyPenalty"]

Results:

• ✅ Auto-titling now works consistently with xAI models
• ✅ No more parameter compatibility issues
• ✅ Clean, descriptive conversation titles generated

Additional Notes:

• The issue appeared to be most prominent when using Grok-4 models specifically
• Switching to grok-3-latest as the title model resolved the stability issues
• The dropParams configuration was key for eliminating the parameter conflicts

Environment Context:

Running on production environment with the configuration working reliably across multiple conversations.

Thanks again for the fast turnaround and clear documentation! The solution works exactly as described. 🚀