最近 ChatGPT 很火，也注册了账号玩了玩，确实灰常强大。但是也有的小伙伴可能没办法注册账号，我就想着把qq群机器人接入ChatGPT。 过程还是比较简单顺利的。下面简单介绍一下
1. ChatGPT 网页的几个接口介绍。
1.1 第一个接口 https://chat.openai.com/backend-api/moderations
1.2 第二个接口 https://chat.openai.com/backend-api/conversation
1.3 第三个接口 https://chat.openai.com/api/auth/session
1.4 总结: 经过我的捣鼓，再加上 github 看了下别人写的插件，可以像下面这样正确的发送请求已经接受响应。
2. ChatGPT 具体请求示例
1. 复制cookie
示例:
"__Secure-next-auth.session-token=eyJhbGciOiJkaXIiLCJlbmMiOiJBMjU2R0NNIn0..R3Kc7SzojKpBGjqD.A8_9DrrtRoHFzEiJXrfWzePQg后面省略..."
ps: (__Secure-next-auth.callback-url 和 __Host-next-auth.csrf-token 可要可不要)
2. 发起session请求，这里用java okhttp实现，没有多余封装，直接硬码
public static String[] sessionReq(String copyCookie) throws IOException {
//copyCookie就是从浏览器请求的请求头，或者 application -> cookie 复制出 __Secure-next-auth.session-token 的key和value。
//ps: (__Secure-next-auth.callback-url 和 __Host-next-auth.csrf-token 可要可不要)
//示例: "__Secure-next-auth.session-token=eyJhbGciOiJkaXIiLCJlbmMiOiJBMjU2R0NNIn0....后面省略"
HashMap<String, String> headerMap = Maps.newHashMap();
headerMap.put("user-agent", "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/107.0.0.0 Safari/537.36");
headerMap.put("cookie", copyCookie);
headerMap.put("accept-encoding", "gzip, deflate, br");
//session接口
Request getRequest = new Request.Builder()
.url("https://chat.openai.com/api/auth/session")
.get()
.headers(Headers.of(headerMap))
.build();
Response responseSession = client.newCall(getRequest).execute();
String result = CharStreams.toString(new InputStreamReader(responseSession.body().byteStream(), StandardCharsets.UTF_8));
logger.info("session response: {}", result);
Map map = objectMapper.readValue(result, Map.class);
String setCookie = responseSession.headers().get("set-cookie");
Map<String, String> collect = Splitter.on(";").splitToList(setCookie)
.stream().filter(it -> it.contains("=")).map(it -> it.split("="))
.collect(Collectors.toMap(it -> it[0], it -> it[1]));
String nextAuthSessionToken = collect.get("__Secure-next-auth.session-token");
String authorization = (String) map.get("accessToken");
logger.info("nextAuthSessionToken: {}", nextAuthSessionToken);
ChatGPTHandle.sessionToken = nextAuthSessionToken;
logger.info("authorization: {}", authorization);
return new String[]{nextAuthSessionToken, authorization};
}
3. 发起 conversation 请求，同样用java okhttp实现，没有多余封装，直接硬码
public static String conversation(String[] auths, String requestBody) throws IOException {
//这个 requestBody 可以作为模板写死，不同的请求只需要修改里面的query
// String requestBody = "{\"parent_message_id\":\"" + UUID.randomUUID()
// + "\",\"action\":\"next\",\"messages\":[{\"role\":\"user\",\"id\":\""
// + UUID.randomUUID() + "\",\"content\":{\"content_type\":\"text\",\"parts\":[\"" + "query" + "\"]}}]," +
// "\"model\":\"text-davinci-002-render\"}";
String nextAuthSessionToken = auths[0];
String authorization = auths[1];
//替换那个 cookie
Map<String, String> conCookieMap = new HashMap<>(4, 1);
//这个就是上面复制的cookie里的内容
conCookieMap.put("__Secure-next-auth.session-token", nextAuthSessionToken);
StringBuilder sb = new StringBuilder();
conCookieMap.forEach((k, v) -> sb.append(k).append("=").append(v).append("; "));
sb.deleteCharAt(sb.length() - 2);
HashMap<String, String> hashMap = Maps.newHashMap();
hashMap.put("accept-encoding", "gzip, deflate, br");
hashMap.put("accept-language", "zh-CN,zh;q=0.9");
hashMap.put("authorization", "Bearer " + authorization);
hashMap.put("content-type", "application/json");
hashMap.put("cookie", sb.toString().trim());
hashMap.put("origin", "https: //chat.openai.com");
hashMap.put("referer", "https: //chat.openai.com/chat");
hashMap.put("user-agent", "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/107.0.0.0 Safari/537.36");
Request post = new Request.Builder()
.headers(Headers.of(hashMap))
.url("https://chat.openai.com/backend-api/conversation")
.post(RequestBody.create(MediaType.parse("application/json"), requestBody)).build();
Call call = client.newCall(post);
Response response = call.execute();
if (response.isSuccessful()) {
//处理response的响应消息
String res = CharStreams.toString(new InputStreamReader(response.body().byteStream(), StandardCharsets.UTF_8));
//这里是连续多行
//a
//a b
//a b c
//这直接取倒数第二行
String[] split = res.split("\n");
List<String> collect1 = Arrays.stream(split).filter(Strings::isNotBlank)
.collect(Collectors.toList());
String fullLine = collect1.get(collect1.size() - 2);
Map map1 = objectMapper.readValue(fullLine.substring(5), Map.class);
ArrayList list = (ArrayList) ((Map) ((Map) map1.get("message")).get("content")).get("parts");
return (String) list.get(0);
} else {
logger.info(response.code() + " " + response.toString());
}
return "服务出错了, g了";
}