titleModel for custom endpoint

[wipash]

What happened?

I have a custom endpoint defined as follows:

endpoints:
  custom:
    - name: LiteLLM
      apiKey: "$${LITELLM_API_KEY}"
      baseURL: "http://litellm.litellm.svc.cluster.local:4000"
      models:
        default: ["azure-gpt-4.1"]
        fetch: true
      titleConvo: true
      titleModel: azure-gpt-4o-mini

When starting a new conversation with a modelSpec or a model directly from this endpoint, LibreChat uses the current model for title generation instead of the defined titleModel.

I have not yet been able to test this against standard endpoints (eg OpenAI directly).

Version Information

ghcr.io/danny-avila/librechat-dev:latest@sha256:a4a51b3cc6e2e0f28060207c1c62b7bd6fe0d30278264cb84d96caa39837dfbe

Steps to Reproduce

Define a custom endpoint as above
Start a new conversation
LiteLLM logs show title request sent to the same model as the chat

What browsers are you seeing the problem on?

No response

Relevant log output

2025-08-12T02:32:09.814Z debug: [BaseClient[] Loading history:
{
  conversationId: "7501f9cd-7dfe-40f8-8005-004bfbef2a89",
  parentMessageId: "00000000-0000-0000-0000-000000000000",
}
2025-08-12T02:32:09.821Z debug: [BaseClient[] Context Count (1/2)
{
  remainingContextTokens: 942806,
  maxContextTokens: 942818,
}
2025-08-12T02:32:09.822Z debug: [BaseClient[] Context Count (2/2)
{
  remainingContextTokens: 942806,
  maxContextTokens: 942818,
}
2025-08-12T02:32:09.823Z debug: [BaseClient[] tokenCountMap:
{
  0a329a51-ce39-4482-ac44-0bef4e907382: 9,
}
2025-08-12T02:32:09.823Z debug: [BaseClient[]
{
  promptTokens: 12,
  remainingContextTokens: 942806,
  payloadSize: 1,
  maxContextTokens: 942818,
}
2025-08-12T02:32:09.835Z debug: [BaseClient[] tokenCountMap
{
  0a329a51-ce39-4482-ac44-0bef4e907382: 9,
}
2025-08-12T02:32:09.836Z debug: [BaseClient[] userMessage
{
  messageId: "0a329a51-ce39-4482-ac44-0bef4e907382",
  parentMessageId: "00000000-0000-0000-0000-000000000000",
  conversationId: "7501f9cd-7dfe-40f8-8005-004bfbef2a89",
  sender: "User",
  text: "Tell me about sun spots",
  isCreatedByUser: true,
  tokenCount: 9,
}
2025-08-12T02:32:09.927Z debug: [saveConvo[] api/app/clients/BaseClient.js - saveMessageToDatabase #saveConvo
2025-08-12T02:32:12.807Z debug: Memory Agent processed memory successfully
["{\"type\":\"text\",\"text\":\"Sunspots are temporary phenomena on the Sun's photosphere that appear as spot... [truncated[]"]
2025-08-12T02:32:15.078Z debug: [spendTokens[] conversationId: 7501f9cd-7dfe-40f8-8005-004bfbef2a89 | Context: message | Token usage:
{
  promptTokens: 12,
  completionTokens: 342,
}
2025-08-12T02:32:15.118Z debug: [spendTokens[] No transactions incurred against balance
2025-08-12T02:32:15.153Z debug: [saveConvo[] api/app/clients/BaseClient.js - saveMessageToDatabase #saveConvo
2025-08-12T02:32:15.190Z debug: [AgentController[] Request closed
2025-08-12T02:32:15.720Z debug: [saveConvo[] api/server/services/Endpoints/agents/title.js
2025-08-12T02:32:15.749Z debug: [AgentController[] Title generation started
2025-08-12T02:32:15.749Z debug: [AgentController[] Title generation completed
2025-08-12T02:32:15.749Z debug: [AgentController[] Performing cleanup
2025-08-12T02:32:15.750Z debug: [AgentController[] Cleaning up abort controller
2025-08-12T02:32:15.750Z debug: [AgentController[] Cleanup completed

LiteLLM request, showing incorrect model (should be azure-gpt-4o-mini):

{
  "user": "678db103cfeexxxxx",
  "model": "azure-gpt-4.1",
  "stream": false,
  "messages": [
    {
      "role": "user",
      "content": "Provide a concise, 5-word-or-less title for the conversation, using title case conventions. Only return the title itself.\n\nConversation:\nUser: Tell me about sun spots\nAI: **Sunspots** are temporary, relatively dark patches on the **Sun’s surface** (photosphere) caused by intense magnetic activity. Here’s an overview:\n\n### What are Sunspots?\n- **Dark, Cooler Areas:** Sunspots appear dark because they're cooler (about 4,000°C) than the surrounding photosphere (about 5,500°C).\n- **Size:** They can be huge—some sunspots are several times larger than Earth.\n- **Lifespan:** Individual sunspots can last from hours to weeks.\n\n### Why do Sunspots Occur?\n- **Magnetic Activity:** Sunspots are regions where the Sun’s magnetic field is extremely strong, which inhibits the flow of hot plasma from the Sun’s interior to the surface, making these regions cooler and thus darker.\n- **Solar Cycle:** The number and size of sunspots fluctuate in roughly 11-year cycles, known as the **solar cycle.** There's ... (litellm_truncated 703 chars)"
    }
  ],
  "max_tokens": 75
}

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

Working as expected here:

    - name: "OpenRouter Gateway"
      apiKey: "${OPENROUTER_KEY}"
      baseURL: "https://gateway.ai.cloudflare.com/v1/id/name/openrouter"
      models:
        default: ["gpt-4o", "gpt-4.1", "google/gemma-3n-e4b-it"]
        fetch: true
      titleConvo: true
      titleModel: "openai/gpt-4.1-nano"
      modelDisplayLabel: "OpenRouter"

Using any model in any custom endpoint has the same effect for me:

[wipash]

Weird, it's the same on two different instances of LibreChat I'm running.
I'll set up OpenRouter and test with that too, in case it's something weird happening at the LiteLLM side.

[wipash]

It's happening with OpenRouter too:

    - name: OpenRouter
      apiKey: "$${OPENROUTER_API_KEY}"
      baseURL: "https://openrouter.ai/api/v1"
      models:
        default: ["openai/gpt-4o"]
        fetch: true
      titleConvo: true
      titleModel: openai/gpt-4.1-nano
      modelDisplayLabel: "OpenRouter"

[wipash]

I tried setting the shared endpoint setting as well, with no change:

endpoints:
  custom:
    - name: LiteLLM
      apiKey: "$${LITELLM_API_KEY}"
      baseURL: "http://litellm.litellm.svc.cluster.local:4000"
      models:
        default: ["azure-gpt-4.1"]
        fetch: true
      titleConvo: true
      titleModel: openai/gpt-4.1-mini

    - name: OpenRouter
      apiKey: "$${OPENROUTER_API_KEY}"
      baseURL: "https://openrouter.ai/api/v1"
      models:
        default: ["openai/gpt-4o"]
        fetch: true
      titleConvo: true
      titleModel: openai/gpt-4.1-nano
      modelDisplayLabel: "OpenRouter"

  assistants:
    disableBuilder: false
  agents:
    recursionLimit: 50
    disableBuilder: false
    capabilities:
      # - "execute_code"
      - "file_search"
      - "actions"
      - "tools"
      - "artifacts"
      - "ocr"
      - "chain"
      - "web_search"

  all:
    streamRate: 35
    titleConvo: true
    titleModel: openai/gpt-4.1-mini
    titleEndpoint: LiteLLM

[Odrec]

Danny's example has the title model between quotes. Long shot but could that be something?