Function parseOpenAIResponse

test/harness/replayingCapiProxy.ts:1258–1289

(
  responseBody: string | undefined,
)

// Takes raw HTTP response data and turns it into an OpenAI ChatCompletion object, regardless of whether
// it's a streaming or non-streaming response
async function parseOpenAIResponse(
  responseBody: string | undefined,
): Promise<ChatCompletion | undefined> {
  // Check if it's a streaming response (Server-Sent Events format)
  if (responseBody?.startsWith("data:")) {
    const lines = responseBody
      .split("\n")
      .filter((line) => line.startsWith("data:") && !line.includes("[DONE]"));
    const chunks = lines.map(
      (line) => JSON.parse(line.slice(5)) as ChatCompletionChunk,
    );

    // Convert the sequence of chunks into a final ChatCompletion object
    // TODO: Do we need to apply fixCAPIStreamingToolCalling normalization here?
    const readableStream = new ReadableStream({
      async start(controller) {
        for (const chunk of chunks) {
          controller.enqueue(
            new TextEncoder().encode(JSON.stringify(chunk) + "\n"),
          );
        }
        controller.close();
      },
    });

    return await ChatCompletionStream.fromReadableStream(
      readableStream,
    ).finalChatCompletion();
  } else if (responseBody) {
    return JSON.parse(responseBody) as ChatCompletion;
  }
}

// Checks if requestMessages is a prefix of savedMessages,
// and returns the index of the next assistant message if found.