Change providers to accept extra parameter

[glGarg]

Adds support for passing custom fields via an extraBody property throughout the BYOK (Bring Your Own Key) model configuration and request pipeline. This enables providers to specify additional request parameters (such as special options for certain endpoints) in a flexible way. The changes ensure that these custom fields are available in model metadata, configuration, and are included in outgoing requests.

[glGarg]

This is how I set the thinking parameter in the config:

customModels:
  customoai:
    accounts/msft/deployments/wxbsoe65:
      name: 'wxbsoe65'
      url: 'https://api.fireworks.ai/inference/v1/chat/completions'
      toolCalling: true
      vision: false
      thinking: false
      maxInputTokens: 128000
      maxOutputTokens: 16384
      requiresAPIKey: true
      extraBody:
        thinking:
          type: disabled

[Copilot]

Pull request overview

Adds support for propagating a provider-specified extraBody object through BYOK model metadata/config so OpenAI-compatible endpoints can inject additional request fields when forming outbound requests.

Changes:

• Extend model metadata/types (IChatModelInformation, BYOKModelCapabilities) to include extraBody.
• Plumb extraBody from Custom OAI model configuration into resolved model info.
• Merge extraBody into the outgoing OpenAI request body in OpenAIEndpoint.

Show a summary per file

File	Description
src/platform/endpoint/common/endpointProvider.ts	Adds extraBody to chat model metadata shape.
src/extension/byok/vscode-node/customOAIProvider.ts	Allows custom OAI model configs to specify extraBody and forwards it into model metadata.
src/extension/byok/node/openAIEndpoint.ts	Merges extraBody into request payloads before sending.
src/extension/byok/common/byokProvider.ts	Carries extraBody through BYOK capability resolution into IChatModelInformation.

Copilot's findings

• Files reviewed: 4/4 changed files
• Comments generated: 2

[Copilot]

Object.assign(body, this.modelMetadata.extraBody) merges unvalidated, user-configurable data into the request body. This can lead to prototype pollution via keys like __proto__, constructor, or prototype, and also produces odd results if extraBody is not a plain object at runtime. Consider validating extraBody is a non-null plain object and filtering out dangerous keys before merging (or doing a safe merge helper).

[Copilot]

New behavior merges modelMetadata.extraBody into outgoing requests, but there are existing unit tests for OpenAIEndpoint and none appear to assert this merge behavior. Please add coverage that verifies extraBody fields are applied (and, if you add validation/sanitization, that unsafe keys are ignored).

[battaglia01]

I was just looking for this feature and am glad it is being added, as we can then use reasoning in OpenRouter.

@glGarg Would we put this in chatLanguageModels.json or somewhere else? The config you have above doesn't look like json.