This SDK simplifies the process of building GitHub Copilot Extensions. Building Copilot Extensions previously required manual handling of request verification, response formatting, and API interactions. This SDK simplifies these tasks, allowing you to focus on your extension's core functionality rather than building boilerplate code. Use it to integrate your tools, APIs, or data sources directly into Copilot Chat.
We consider this SDK alpha software in terms of API stability, but we adhere to semantic-versioning, so it's safe to use today.
- Request payload verification
- Payload parsing
- Response building
- Handles security and response formatting requirements
- Provides utilities for common extension tasks
- Streamlines the development process
import { verifyRequestByKeyId } from "@copilot-extensions/preview-sdk";
const { isValid, cache } = await verifyRequestByKeyId(
request.body,
signature,
keyId,
{
token: process.env.GITHUB_TOKEN,
},
);
// isValid: true or false
// cache: { id, keys }import { createAckEvent, createDoneEvent, createTextEvent } from "@copilot-extensions/preview-sdk";
export default handler(request, response) {
const ackEvent = createAckEvent();
const textEvent = createTextEvent("Hello, world!");
const doneEvent = createDoneEvent();
response.write(ackEvent);
response.write(textEvent);
response.end(doneEvent);
}import { prompt } from "@copilot-extensions/preview-sdk";
try {
const { message } = await prompt("What is the capital of France?", {
token: process.env.TOKEN,
});
console.log(message.content);
} catch (error) {
console.error(error);
}Verify the request payload using the provided signature and key ID. The method will request the public key from GitHub's API for the given keyId and then verify the payload.
The requestOptions argument is optional. It can contain:
- a
tokento authenticate the request to GitHub's API - a custom octokit
requestinstance to use for the request - a
cacheto use cached keys
import { verifyRequestByKeyId } from "@copilot-extensions/preview-sdk";
const { isValid, cache } = await verifyRequestByKeyId(
request.body,
signature,
keyId,
);
// with token
const { isValid, cache } = await verifyRequestByKeyId(
request.body,
signature,
keyId,
{ token: "ghp_1234" },
);
// with custom octokit request instance
const { isValid, cache } = await verifyRequestByKeyId(
request.body,
signature,
keyId,
{ request },
);
// with cache
const previousCache = {
id: "etag_value",
keys: [{ key_identifier: "key1", key: "public_key1", is_current: true }],
};
const { isValid, cache } = await verifyRequestByKeyId(
request.body,
signature,
keyId,
{ cache: previousCache },
);Fetches public keys for verifying copilot extension requests from GitHub's API and returns them as an array. The request can be made without authentication, with a token, with a custom octokit request instance, or with a cache.
import { fetchVerificationKeys } from "@copilot-extensions/preview-sdk";
// fetch without authentication
const { id, keys } = await fetchVerificationKeys();
// with token
const { id, keys } = await fetchVerificationKeys({ token: "ghp_1234" });
// with custom octokit request instance
const { id, keys } = await fetchVerificationKeys({ request });
// with cache
const cache = {
id: "etag_value",
keys: [{ key_identifier: "key1", key: "public_key1" }],
};
const { id, keys } = await fetchVerificationKeys({ cache });Verify the request payload using the provided signature and key. Note that the raw body as received by GitHub must be passed, before any parsing.
import { verify } from "@copilot-extensions/preview-sdk";
const payloadIsVerified = await verifyRequest(request.body, signature, key);
// true or falseAll create*Event() methods return a string that can directly be written to the response stream.
Acknowledge the request so that the chat UI can tell the user that the agent started generating a response.
The ack event should only be sent once.
import { createAckEvent } from "@copilot-extensions/preview-sdk";
response.write(createAckEvent());Send a text message to the chat UI. Multiple messages can be sent. The message argument must be a string and may include markdown.
import { createTextEvent } from "@copilot-extensions/preview-sdk";
response.write(createTextEvent("Hello, world!"));
response.write(createTextEvent("Hello, again!"));Ask the user to confirm an action. The confirmation event should only be sent once.
The meta data object will be sent along the user's response.
See additional documentation about Copilot confirmations.
import { createConfirmationEvent } from "@copilot-extensions/preview-sdk";
response.write(
createConfirmationEvent({
id: "123",
title: "Are you sure?",
message: "This will do something.",
}),
);Send a list of references to the chat UI. The references argument must be an array of objects with the following properties:
idtype
The following properties are optional
data: object with any propertiesis_implicit: a booleanmetadata: an object with a requireddisplay_nameand the optional properties:display_iconanddisplay_url
Multiple references events can be sent.
See additional documentation about Copilot references.
import { createReferencesEvent } from "@copilot-extensions/preview-sdk";
response.write(
createReferencesEvent([
{
id: "123",
type: "issue",
data: {
number: 123,
},
is_implicit: false,
metadata: {
display_name: "My issue",
display_icon: "issue-opened",
display_url: "https://github.com/monalisa/hello-world/issues/123",
},
])
);An array of objects with the following properties:
type: must be one of:"reference","function","agent"codemessageidentifier
See additional documentation about Copilot errors.
The done event should only be sent once, at the end of the response. No further events can be sent after the done event.
import { createDoneEvent } from "@copilot-extensions/preview-sdk";
response.write(createDoneEvent());Parses the raw request body and returns an object with type support.
index.d.ts to improve it
import { parseRequestBody } from "@copilot-extensions/preview-sdk";
const payload = parseRequestBody(rawBody);
// When your IDE supports types, typing "payload." should prompt the available keys and their types.For cases when you want to pipe a user request directly to OpenAI, use this method to remove Copilot-specific fields from the request payload.
import { transformPayloadForOpenAICompatibility } from "@copilot-extensions/preview-sdk";
import { OpenAI } from "openai";
const openaiPayload = transformPayloadForOpenAICompatibility(payload);
const openai = new OpenAI({ apiKey: OPENAI_API_KEY });
const stream = openai.beta.chat.completions.stream({
...openaiPayload,
model: "gpt-4-1106-preview",
stream: true,
});Convenience method to verify and parse a request in one go. It calls verifyRequestByKeyId() and parseRequestBody() internally.
import { verifyAndParseRequest } from "@copilot-extensions/preview-sdk";
const { isValidRequest, payload, cache } = await verifyAndParseRequest(
request.body,
signature,
keyId,
{
token: process.env.GITHUB_TOKEN,
},
);
if (!isValidRequest) {
throw new Error("Request could not be verified");
}
// `isValidRequest` is a boolean.
// `payload` has type support.
// `cache` contains the id and keys used for verification.Convenience method to get the user's message from the request payload.
import { getUserMessage } from "@copilot-extensions/preview-sdk";
const userMessage = getUserMessage(payload);Convenience method to get the user's confirmation from the request payload (in case the user's last response was a confirmation).
import { getUserConfirmation } from "@copilot-extensions/preview-sdk";
const userConfirmation = getUserConfirmation(payload);
if (userConfirmation) {
console.log("Received a user confirmation", userConfirmation);
} else {
// The user's last response was not a confirmation
}Send a prompt to the chat UI and receive a response from the user. The message argument must be a string and may include markdown.
The options argument is optional. It can contain a token to authenticate the request to GitHub's API, or a custom request.fetch instance to use for the request.
import { prompt } from "@copilot-extensions/preview-sdk";
const { message } = await prompt("What is the capital of France?", {
model: "gpt-4",
token: process.env.TOKEN,
});
console.log(message.content);In order to pass a history of messages, pass them as options.messages:
const { message } = await prompt("What about Spain?", {
model: "gpt-4",
token: process.env.TOKEN,
messages: [
{ role: "user", content: "What is the capital of France?" },
{ role: "assistant", content: "The capital of France is Paris." },
],
});Alternatively, skip the message argument and pass all messages as options.messages:
const { message } = await prompt({
model: "gpt-4",
token: process.env.TOKEN,
messages: [
{ role: "user", content: "What is the capital of France?" },
{ role: "assistant", content: "The capital of France is Paris." },
{ role: "user", content: "What about Spain?" },
],
});await prompt({
model: "gpt-4o",
token: process.env.TOKEN,
system: "You are a helpful assistant.",
messages: [
{ role: "user", content: "What is the capital of France?" },
{ role: "assistant", content: "The capital of France is Paris." },
{
role: "user",
content: [
[
{ type: "text", text: "What about this country?" },
{
type: "image_url",
image_url: urlToImageOfFlagOfSpain,
},
],
],
},
],
// GitHub recommends using your GitHub username, the name of your application
// https://docs.github.com/en/rest/using-the-rest-api/getting-started-with-the-rest-api?apiVersion=2022-11-28#user-agent
userAgent: "gr2m/my-app v1.2.3",
// set an alternative chat completions endpoint
endpoint: "https://models.inference.ai.azure.com/chat/completions",
// compare https://platform.openai.com/docs/guides/function-calling/configuring-function-calling-behavior-using-the-tool_choice-parameter
toolChoice: "auto",
tools: [
{
type: "function",
function: {
name: "get_weather",
strict: true,
parameters: {
type: "object",
properties: {
location: { type: "string" },
unit: { type: "string", enum: ["c", "f"] },
},
required: ["location", "unit"],
additionalProperties: false,
},
},
},
],
// configuration related to the request transport layer
request: {
// for mocking, proxying, client certificates, etc.
fetch: myCustomFetch,
// hook into request life cycle for complex authentication strategies, retries, throttling, etc
// compare options.request.hook from https://github.com/octokit/request.js
hook: myCustomHook,
// Use an `AbortController` instance to cancel a request
signal: myAbortController.signal,
},
});Works the same way as prompt(), but resolves with a stream key instead of a message key.
import { prompt } from "@copilot-extensions/preview-sdk";
const { requestId, stream } = await prompt.stream("What is the capital of France?", {
token: process.env.TOKEN,
});
for await (const chunk of stream) {
console.log(new TextDecoder().decode(chunk));
}Convenience metthod if a result from a prompt() call includes function calls.
import { prompt, getFunctionCalls } from "@copilot-extensions/preview-sdk";
const result = await prompt(options);
const [functionCall] = getFunctionCalls(result);
if (functionCall) {
console.log("Received a function call", functionCall);
} else {
console.log("No function call received");
}- Using Copilot Extensions
- About building Copilot Extensions
- Set up process
- Communicating with the Copilot platform
- Communicating with GitHub
While implementing the lower-level functionality, we also dream big: what would our dream SDK for Coplitot extensions look like? Please have a look and share your thoughts and ideas:
Please see CONTRIBUTING.md.