CopilotBackend.mdx

title	CopilotBackend
description	Handles requests from frontend, provides function calling and various LLM backends.

```jsx CopilotKitBackend Example import { CopilotBackend, OpenAIAdapter } from "@copilotkit/backend";

export async function POST(req: Request) { const copilotKit = new CopilotBackend(); return copilotKit.response(req, new OpenAIAdapter()); }

</RequestExample>

This class is the main entry point for the backend. It handles requests from the frontend, provides function calling and various LLM backends.

Currently, you can use `OpenAIAdapter` for direct access to the OpenAI API and `LangChainAdapter` to use the LangChain API as a backend.

## OpenAIAdapter

Simply pass an instance of `OpenAIAdapter` to the `response` method of `CopilotBackend` to use OpenAI as a backend.

```typescript
const copilotKit = new CopilotBackend();
return copilotKit.response(req, new OpenAIAdapter());

LangChainAdapter

To use LangChain as a backend, provide a handler function to the adapter with your custom LangChain logic. The async handler function can return:

• a simple string response
• a LangChain stream IterableReadableStream
• a LangChain BaseMessageChunk object
• a LangChain AIMessage object

This example streams back OpenAI messages via LangChain:

return copilotKit.response(
  req,
  new LangChainAdapter(async (forwardedProps) => {
    const model = new ChatOpenAI(
      { modelName: "gpt-4-1106-preview" }
    );
    return model.stream(
      forwardedProps.messages, 
      { tools: forwardedProps.tools }
    );
  }),
);

Server Side Actions

CopilotKit supports actions that can be executed on the server side. You can define server side actions by passing the actions parameter:

const copilotKit = new CopilotBackend({
  actions: [
    {
      name: "sayHello",
      description: "Says hello to someone.",
      argumentAnnotations: [
        {
          name: "arg",
          type: "string",
          description: "The name of the person to say hello to.",
          required: true,
        },
      ],
      implementation: async (arg) => {
        console.log("Hello from the server", arg, "!");
      },
    },
  ],
});

Server side actions can also return a result which becomes part of the message history.

This is useful because it gives the LLM context about what happened on the server side. In addition, it can be used to look up information from a vector or relational database and other sources.

In addition to that, server side actions can also come from LangChain, including support for streaming responses.

Returned results can be of the following type:

• anything serializable to JSON
• string
• LangChain types:
  • IterableReadableStream
  • BaseMessageChunk
  • AIMessage

LangServe

The backend also supports LangServe, enabling you to connect to existing chains, for example python based chains. Use the langserve parameter to specify URLs for LangServe.

const copilotKit = new CopilotBackend({
  langserve: [
    {
      chainUrl: "http://my-langserve.chain",
      name: "performResearch",
      description: "Performs research on a given topic.",
    }
  ],
});

When left out, arguments are automatically inferred from the schema provided by LangServe.

Constructor

A list of server side actions that can be executed. An array of LangServer URLs.

response(request, adapter)

Handle a HTTP request and return a Response object.

The HTTP request object. The adapter to use for the response.

stream(forwardedProps, adapter)

Returns a ReadableStream that can be used to stream messages back to the client.

The properties forwarded to the adapter. This includes messages, actions etc. The adapter to use for the response.

streamHttpServerResponse(req, res, adapter)

Streams messages back to the client using the HTTP response object. Use with express, or Node.js HTTP server.

The HTTP request object. The HTTP response object. The adapter to use for the response.