Kirara AI
lss233/kirara-aiReviewed by AI
Checked by RepoPicks Radar on
What is Kirara AI?
Kirara AI puts an AI chatbot inside chat apps. You run it on your own machine or server, pick a model (OpenAI, DeepSeek, Claude, Gemini, Qwen, Doubao, Ollama and others), connect it to QQ, Telegram, WeChat or Discord, and set personalities, keyword triggers, workflows, image generation and voice replies from a web admin page. It is aimed at people who like tinkering: the project manual and the community are in Chinese, and you install it with Python or Docker. A plugin market adds more features.
Kirara AI is free and open source (AGPL-3.0) and can replace Chatbase.
How easy is it to install Kirara AI?
C · install 1 or 2 tools · Windows
- Windows C
- macOS D
- Linux D
Windows is the easiest: there is a ready "Windows-quickstart" zip, but it belongs to a pre-release from May 2025, so for the current version you install Python and the kirara-ai package and then connect a model and a chat account in the web admin. On Linux and macOS the route is Docker (the repository ships a Dockerfile and the project publishes an image), plus the same model and chat-platform setup, and the manual is in Chinese.
Kirara AI on GitHub
- Stars
- 19,277on Nov 10, 2026
- Downloads
- 129,040all releases
- Growth
- +14stars in 6 days
Stars over time
Facts
- Runs on
- Windows, Linux, macOS
- You need to install
- Python, Docker
- To build from source
- Python, Docker
- Disk space
- 219 MB–657 MB · from the installer
- Accounts or keys
- An AI model provider (OpenAI, DeepSeek, Claude, Gemini, Qwen and others), A chat platform bot token (QQ, Telegram, WeChat, Discord), Azure Speech (voice replies), Midjourney or Stable Diffusion service (image replies)
- Cost
- Free · The program is free (AGPL-3.0). You pay whatever the AI model costs: the paid providers charge per use, and a local model through Ollama costs nothing. Voice replies through Azure and image models like Midjourney are paid separately.
- License
- AGPL-3.0 (Copyleft)
- Can replace
- ChatGPT Plus, Chatbase
- Last change on GitHub
- Oct 22, 2026
Kirara AI risk signals (6)
- Attentionconnects your personal chat accounts
The bot plugs into chat platforms that do not allow third-party clients for personal accounts: the repository topics include "go-cqhttp", "mirai", "qq" and "wechat", the dependency list includes "wechatpy", and the platform table lists "OneBot" adapters. Tencent's rules for QQ and WeChat forbid unofficial clients, and accounts used this way can be blocked.
AI review
- Infowindows package is a prerelease
The newest release, v3.2.1 from 2026-10-22, has "no platform binaries". The only Windows package is "Windows-quickstart-kirara-ai-v3.3.0a2.zip" from 2025-05-05, an alpha version (pyproject.toml says version = "3.3.0a2").
AI review
- Infointerface downloaded at build time
The Dockerfile does not build the web interface from this repository; it fetches whatever is newest from npm while building: "PACKAGE_INFO=$(curl -s https://registry.npmjs.org/kirara-ai-webui) && LATEST_VERSION=$(printf %s $PACKAGE_INFO | jq -r '.["dist-tags"].latest')". The interface code is therefore not pinned to the version you checked.
AI review
- Infopackages from a mirror
pyproject.toml makes a mirror in China the default source for Python packages: "[[tool.uv.index]] url = \"https://mirrors.ustc.edu.cn/pypi/simple\" default = true".
AI review
- Infobuiltin tunnel feature
The feature list includes a built-in tunnel that makes the local service reachable from outside your network: "内置 Frpc 内网穿透" (built-in Frpc tunneling). The project also exposes an HTTP API with /v1/chat and /v2/chat endpoints.
AI review
- Infodocumentation in chinese
The README, the setup guide link ("请移步至 快速开始"), the manual at kirara-docs.app.lss233.com and the community groups are in Chinese; there is no English README in the repository.
AI review
Evidence is shown in its original language.
What Radar could not check about Kirara AI
- Whether the 2025 Windows quickstart zip still works with the current plugins and models
- Whether the QQ, WeChat and Discord adapters work today (the table marks Discord and Feishu as being rebuilt)
- How much the chosen AI model costs in a busy group chat
- What the plugins from the plugin market do once installed
- Real disk use after the Docker image and the models are in place