You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

4.7 KiB

Raw Blame History

Conversation API Implementation Logic

The implementation of LobeChat's large model AI mainly relies on OpenAI's API, including the core conversation API on the backend and the integrated API on the frontend. Next, we will introduce the implementation approach and code for the backend and frontend separately.

Backend Implementation

The following code removes authentication, error handling, and other logic, retaining only the core functionality logic.

Core Conversation API

In the file src/app/api/openai/chat/handler.ts, we define a POST method, which first parses the payload data from the request (i.e., the conversation content sent by the client), and then retrieves the authorization information from the request. Then, we create an openai object and call the createChatCompletion method, which is responsible for sending the conversation request to OpenAI and returning the result.

export const POST = async (req: Request) => {
  const payload = await req.json();

  const { apiKey, endpoint } = getOpenAIAuthFromRequest(req);

  const openai = createOpenai(apiKey, endpoint);

  return createChatCompletion({ openai, payload });
};

Conversation Result Processing

In the file src/app/api/openai/chat/createChatCompletion.ts, we define the createChatCompletion method, which first preprocesses the payload data, then calls OpenAI's chat.completions.create method to send the request, and uses the OpenAIStream from the Vercel AI SDK to convert the returned result into a streaming response.

import { OpenAIStream, StreamingTextResponse } from 'ai';

export const createChatCompletion = async ({ payload, openai }: CreateChatCompletionOptions) => {
  const { messages, ...params } = payload;

  const formatMessages = messages.map((m) => ({
    content: m.content,
    name: m.name,
    role: m.role,
  }));

  const response = await openai.chat.completions.create(
    {
      messages: formatMessages,
      ...params,
      stream: true,
    },
    { headers: { Accept: '*/*' } },
  );
  const stream = OpenAIStream(response);
  return new StreamingTextResponse(stream);
};

Frontend Implementation

Frontend Integration

In the src/services/chatModel.ts file, we define the fetchChatModel method, which first preprocesses the payload data, then sends a POST request to the /chat endpoint on the backend, and returns the request result.

export const fetchChatModel = (
  { plugins: enabledPlugins, ...params }: Partial<OpenAIStreamPayload>,
  options?: FetchChatModelOptions,
) => {
  const payload = merge(
    {
      model: initialLobeAgentConfig.model,
      stream: true,
      ...initialLobeAgentConfig.params,
    },
    params,
  );

  const filterFunctions: ChatCompletionFunctions[] = pluginSelectors.enabledSchema(enabledPlugins)(
    usePluginStore.getState(),
  );

  const functions = filterFunctions.length === 0 ? undefined : filterFunctions;

  return fetch(OPENAI_URLS.chat, {
    body: JSON.stringify({ ...payload, functions }),
    headers: createHeaderWithOpenAI({ 'Content-Type': 'application/json' }),
    method: 'POST',
    signal: options?.signal,
  });
};

Using Streaming to Get Results

In the src/utils/fetch.ts file, we define the fetchSSE method, which uses a streaming approach to retrieve data. When a new data chunk is read, it calls the onMessageHandle callback function to process the data chunk, achieving a typewriter-like output effect.

export const fetchSSE = async (fetchFn: () => Promise<Response>, options: FetchSSEOptions = {}) => {
  const response = await fetchFn();

  if (!response.ok) {
    const chatMessageError = await getMessageError(response);

    options.onErrorHandle?.(chatMessageError);
    return;
  }

  const returnRes = response.clone();

  const data = response.body;

  if (!data) return;

  const reader = data.getReader();
  const decoder = new TextDecoder();

  let done = false;

  while (!done) {
    const { value, done: doneReading } = await reader.read();
    done = doneReading;
    const chunkValue = decoder.decode(value);

    options.onMessageHandle?.(chunkValue);
  }

  return returnRes;
};

The above is the core implementation of the LobeChat session API. With an understanding of these core codes, further expansion and optimization of LobeChat's AI functionality can be achieved.

4.7 KiB Raw Blame History