Support OpenAI ChatCompletion/Responses store property

[sebastiansandqvist]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

It's not currently possible to enable the store property on chat completions and responses using the OpenAIModel. See types.

This feature request is pretty minimal--to add support for this property:

const openaiModel = new ChatOpenAI({
  model: '...',
+ store: true,
});

• https://platform.openai.com/docs/api-reference/chat/create#chat-create-store

Motivation

Adding store will enable observability on the openai platform.

Proposal (If applicable)

Probably something like this in this file:

export declare interface OpenAIBaseInput {
  /** Sampling temperature to use */
  temperature: number;

  /**
   * Maximum number of tokens to generate in the completion. -1 returns as many
   * tokens as possible given the prompt and the model's maximum context size.
   */
  maxTokens?: number;

  /**
   * Maximum number of tokens to generate in the completion. -1 returns as many
   * tokens as possible given the prompt and the model's maximum context size.
   * Alias for `maxTokens` for reasoning models.
   */
  maxCompletionTokens?: number;

  /** Total probability mass of tokens to consider at each step */
  topP: number;

  /** Penalizes repeated tokens according to frequency */
  frequencyPenalty: number;

  /** Penalizes repeated tokens */
  presencePenalty: number;

  /** Number of completions to generate for each prompt */
  n: number;

  /** Dictionary used to adjust the probability of specific tokens being generated */
  logitBias?: Record<string, number>;

  /** Unique string identifier representing your end-user, which can help OpenAI to monitor and detect abuse. */
  user?: string;

  /** Whether to stream the results or not. Enabling disables tokenUsage reporting */
  streaming: boolean;

  /**
   * Whether or not to include token usage data in streamed chunks.
   * @default true
   */
  streamUsage?: boolean;

  /**
   * Model name to use
   * Alias for `model`
   * @deprecated Use "model" instead.
   */
  modelName: string;

  /** Model name to use */
  model: string;

  /** Holds any additional parameters that are valid to pass to {@link
   * https://platform.openai.com/docs/api-reference/completions/create |
   * `openai.createCompletion`} that are not explicitly specified on this class.
   */
  // eslint-disable-next-line @typescript-eslint/no-explicit-any
  modelKwargs?: Record<string, any>;

  /**
   * List of stop words to use when generating
   * Alias for `stopSequences`
   */
  stop?: string[];
  /** List of stop words to use when generating */
  stopSequences?: string[];

+ /**
+ * Whether or not to store the output of this chat completion request for use in OpenAI's model distillation or evals products.
+ */
+ store?: boolean;

  /**
   * Timeout to use when making requests to OpenAI.
   */
  timeout?: number;

  /**
   * API key to use when making requests to OpenAI. Defaults to the value of
   * `OPENAI_API_KEY` environment variable.
   * Alias for `apiKey`
   */
  openAIApiKey?: string;
  /**
   * API key to use when making requests to OpenAI. Defaults to the value of
   * `OPENAI_API_KEY` environment variable.
   */
  apiKey?: string;
}

With similar updates around these lines.

[kvnn]

~Is there a passthrough / override mechanism, so that any new provider parameters can be used without type additions? ~

Similarly, for OpenAI, metadata should be supported.

@sebastiansandqvist see modelKwargs :

  /** Holds any additional parameters that are valid to pass to {@link
   * https://platform.openai.com/docs/api-reference/completions/create |
   * `openai.createCompletion`} that are not explicitly specified on this class.
   */
  // eslint-disable-next-line @typescript-eslint/no-explicit-any
  modelKwargs?: Record<string, any>;

https://github.com/langchain-ai/langchainjs/blob/main/libs/langchain-openai/src/types.ts#L71C1-L76C37

[sebastiansandqvist]

Thanks @kvnn, that should do the trick for our use case!