go-cqhttp (Mrs4s/go-cqhttp)
Searching Database…
What is this?
A Go implementation of the CQHTTP (OneBot) bot protocol for QQ, built on Mirai and MiraiGo — a lightweight, natively cross-platform bridge that lets you build QQ bots.
Is it a library I import, or a program I run on its own?
It is a standalone program, not a library: it connects to QQ itself and serves your bot through HTTP API, reverse HTTP POST, forward WebSocket and reverse WebSocket endpoints.
How do I get it running?
Download a prebuilt release binary for your platform or use the project's Docker image, run it, and log in with your QQ account. After that it exposes the OneBot endpoints your bot connects to.
Do I need a QQ account for it to work?
Yes — it signs into QQ with a bot account, which is how messages flow. Your own code only needs an HTTP or WebSocket client to call its API.
Is it still actively maintained?
No. The authors state QQ keeps cracking down on protocol libraries with new encryption, so they can no longer maintain go-cqhttp and advise bot developers to migrate to headless NTQQ projects soon.
Will my existing OneBot v11 bot code work with it?
Yes — it is compatible with most of OneBot v11 and adds extensions on top: multi-point HTTP POST reporting, multiple reverse WebSocket connections, group renames, message recall events, and parsing or sending replies and merged forwards.
What can it actually send and receive in QQ?
CQ codes cover QQ faces, voice, short video, @mentions, replies, merged forwards, XML and JSON messages, plus extensions for images, red envelopes, pokes, card images and text-to-speech. Events include private and group messages, member joins and leaves, mutes, recalls, pokes, and friend or group join requests.
What can it do beyond messaging?
OneBot and extension APIs handle group management — kicking or banning members, setting admins, member cards, group names and special titles — plus friend and group list lookups, group file system access, image OCR and Chinese word segmentation.
How much memory does it need?
Very little: with the database off it used about 15MB of RAM after 24 hours with 25 friends and 128 groups. The database adds 10-20MB, so on systems with under 128MB of RAM the authors recommend running without it.
What are similar projects?
kirara-ai, NapCatQQ, openclaw-china, wechaty, openclaw-china-docker, qq-bridge, cc-connect, EverydayWechat