Tianji is a Python-based project focused on applying large language models to social etiquette domains, offering prompt engineering, RAG knowledge bases, agents, and fine-tuning tutorials. It provides multiple integration paths and data tooling; latest activity includes 2025-04-29 push. Repository has 1815 stars and 4 open issues.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Tianji is a large language model application focused on traditional social etiquette scenarios. It covers prompt engineering, RAG knowledge bases, Agent implementations, and model fine-tuning tutorials. The project includes data resources and tutorials for end-to-end LLM development in its domain.
How it works
The project combines several components:
- Prompt engineering modules and demonstrations (prompt version paths)
- RAG knowledge base with LangChain/LlamaIndex-based systems for knowledge-based conversations
- Agent implementations using frameworks like MetaGPT for tool usage
- Finetuning workflows using Transformers and Xtuner for LoRA and full-model fine-tuning
- Data generation and cleaning tools for social etiquette corpora
These components appear to be organized under directories such as tianji/ (core logic), tianji/knowledges (knowledge base), tianji/finetune (finetuning), tools (data tooling), and run (frontends).
Getting started
Quickstart guide is provided in the README:
- Environment installation:
pip install -e .
- Key configuration:
OPENAI_API_KEY=
ZHIPUAI_API_KEY=
HF_HOME='temp/huggingface_cache/'
HF_ENDPOINT='https://hf-mirror.com'
HF_TOKEN=
- Running components (need to create a .env with keys):
# 运行prompt webui前端
python3 run/tianji_prompt_webui.py
# 运行agent前端
streamlit run run/metagpt_webui.py
# 运行langchain前端
python run/demo_rag_langchain_onlinellm.py
Recent releases
Latest RELEASES entry shows none for version 0. The README mentions updates such as the 2024-12-21 Wish model release and various data/model sharing efforts, but there is no formal latest release entry listed in the provided data.
Traction
- stars: 1815
- forks: 157
- open_issues: 4
Behind the repo
- The project references external resources and studios, including a link to a knowledge base and prompt engineering studio on ModelScope, plus HuggingFace datasets for Tianji Chinese data. It also points to a SocialAI ecosystem site for more information. No company incubator link is provided in the given data.
Caveats
- license: Apache-2.0
- created: 2023-12-14
- last_push: 2025-04-29
- language: Python
- There are multiple external dependencies and environment keys required (OPENAI_API_KEY, ZHIPUAI_API_KEY, HF_TOKEN, TAVILY_API_KEY) and a .env file setup is required for running prompts, agents, and RAG demos.