Setting Default Temperature and Range for non-Custom Endpoints

[jcbartle]

I suspect I'm doing something wrong, but I can't get this to work and I can't find an answer:

Is it possible to control the default temperature and range for non-custom endpoints as you can for custom endpoints?

Trying this in the configuration doesn't throw errors - the config file is determined to be valid - but the settings also do not work:

azureOpenAI:
  customParams:
        paramDefinitions:
          - key: temperature
            range:
              min: 0.0
              max: 1.0
              step: 0.1
            default: 0.0

If this is intentionally not supported, is there any mechanism for setting default temperature for non-custom models outside of editing schemas.ts?

[frenzybiscuit]

I would also like to know this!

I tried to follow that exact guide and it doesn't work for me either.

[danny-avila]

customParams is for custom endpoints only, not azureOpenAI

• https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/custom_params

Shared endpoint settings are listed here:

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

[frenzybiscuit]

This doesn't work (it should be showing up for the end user on their panel, right? Or is it just the default that gets sent unless the end user changes it?):

  custom:
    - name: "Ready.Art"
      apiKey: "keyhere"
      baseURL: "https://urlhere.tld/v1"
      models:
        default: ["Darkhn-L3-3-70B-Animus-V7.0-5.25bpw-EXL2"]
        fetch: true
      titleConvo: true
      titleModel: "current_model"
      summarize: false
      summaryModel: ""
      forcePrompt: false
      modelDisplayLabel: "Ready.Art"
      dropParms: ["presence_penalty"]
      customParams:
        paramDefinitions:
          - key: temperature
            range:
              min: 0.0
              max: 1.0
              step: 0.1
            default: 0.0```

[jcbartle]

Hi, @danny-avila. Thanks for the response, but the shared endpoint configurations do not address the need.

cc @nhtruong, as they where critical in getting the custom endpoint custom parameters feature implemented.

In the schemas.ts file, there are a bunch of hardcoded values for the endpoint settings, such as this:

export const openAISettings = {
  model: {
    default: 'gpt-4o-mini' as const,
  },
  temperature: {
    min: 0 as const,
    max: 2 as const,
    step: 0.01 as const,
    default: 1 as const,
  },
  ...
}

Is there any specific reason these could not be configurable from the librechat.yaml file rather than hardcoded? Either that or support customParams for the "standard" endpoints?

In addition to settings like the above, it would be great if we were able to disable certain items from being changed by users, such as the Use Responses API and Web Search settings (we'd like to leave these permanently disabled). Referencing separate conversation on that topic, since it's very related.

Thank you for the consideration!

[oliver-31415]

I have the opposite issue. I don't want to have to select a preset every time I use a model and I want reasoning summaries by default. I was really hoping for a way to have this selected as default. It feels like a missing opportunity for the user to be able to save a setting like that as a default for themselves without having to make a whole new preset. I have tried:
endpoints:
openAI:
summaryModel: "gpt-5-nano"
customParams:
defaultParamsEndpoint: "openAI"
paramDefinitions:
- key: "useResponsesApi"
default: true

to get responses api to be default but have found no way of doing so. This has become pretty standard for a lot of models and you miss out on functionality without responses api. I really hope there is some way of doing this.

[danny-avila]

@oliver-31415 you can use a default preset, or a default modelSpec:

Default preset (old docs, but still relevant):
https://www.librechat.ai/docs/user_guides/presets#setting-default-preset

Model Specs (can set a preset as default):
https://www.librechat.ai/docs/configuration/librechat_yaml/object_structure/model_specs#default

[oliver-31415]

@danny-avila I have done this before, but the thing is I want to use many different models, all with similar settings. I want all of them to have responses api on. This isn't really very easy to do with model specs or presets. I have saved many presets, but I find that they are harder to find the one I want. Keeping the structure of enpoints and models but able to have the settings I want in there would be amazing. I get that this is not currently built in, but I believe that it would improve the functionality. Could I submit a feature request?