[Bug]: Wrong endpoint selected if modelspec options are used

[xhejtman]

What happened?

If I configure model spec such as:

modelSpecs:
      enforce: false
      prioritize: false
      list:
       - name: llama3.3:latest
         label: "Llama 3.3"
         default: true
         iconURL: https://cdn1.cloud.e-infra.cz/logos/meta-llama.png
         showIconInMenu: true
         preset:
           endpoint: "custom"
           model: "llama3.3:latest"
           maxContextTokens: 128000
           max_tokens: 16384

and do not set ENDPOINTS variable (which seems to be optional), and define endpoints in the librechat.yaml:

endpoints:
      agents:
        recursionLimit: 50
        disableBuilder: false
        capabilities: ["file_search", "actions", "tools"]
      custom:
      - name: "Ollama"
        apiKey: "ollama"
        baseURL: "http://ollama:11434/v1/"
        models:
          default: [
            "llama3.3:latest",
            "qwq",
            "qwen2.5-coder"
            ]
          fetch: false # fetching list of models is not supported
        titleConvo: true
        titleModel: "current_model"

the UI "pre-select" Agent endpoint and it cannot be changed to the Ollama endpoint.

It can be fixed setting the ENDPOINTS to "custom,agents".

Cannot be changed = you can click at ollama, but it retuns back to Agents.

Steps to Reproduce

Configure librechat as above.

What browsers are you seeing the problem on?

Chrome

Relevant log output

No logs.

Screenshots

No response

Code of Conduct

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

[danny-avila]

Thanks for your report. There are some issues with model specs, resolving them here: #5218

However, the main issue would be here: endpoint: "custom"; custom is not a valid value for endpoint, it needs to match the actual custom endpoint name.

modelSpecs:
    enforce: false
    prioritize: false
    list:
    - name: llama3.3:latest
      label: "Llama 3.3"
      iconURL: https://cdn1.cloud.e-infra.cz/logos/meta-llama.png
      showIconInMenu: true
      default: true
      preset:
        endpoint: "ollama" # corrected endpoint
        model: "llama3.3:latest"
        maxContextTokens: 128000
        max_tokens: 16384

Right now, it's case-sensitive and default specs are not being applied as you observed in #5214

[danny-avila]

I'll be updating documentation with these notes