ChatDKU

ChatDKU is an agentic RAG system with a web frontend, a Flask backend, and optional Django services. This repository is open-source and designed to run on local machines or servers.

Quickstart (Docker)

1. Copy the environment template:
   • cp .env.example .env
2. Start the stack:
   • docker compose up --build
3. Open the app:
   • http://localhost:3005

To use a local secrets file without committing it, create .env.local and run:

docker compose --env-file .env.local up --build

Environment Configuration

Key variables (see .env.example):

• Core: LLM_API_KEY, LLM_BASE_URL, TEI_URL, EMBEDDING_MODEL
• Storage: REDIS_HOST, REDIS_PASSWORD, CHROMA_HOST, CHROMA_DB_PORT
• Frontend: NEXT_PUBLIC_API_BASE_URL, NEXT_PUBLIC_DICTATION_WS_URL
• Next.js server proxy: BACKEND_INTERNAL_URL, BACKEND_FEEDBACK_URL

LLM / Embedding Services

• LLM: OpenAI-compatible server (sglang). Default guide uses port 18085, so set LLM_BASE_URL=http://<server-ip>:18085/v1 and LLM_API_KEY.
• Embedding: TEI + bge-m3 on port 8080, so set TEI_URL=http://<server-ip>:8080.
• Full server setup: see Documentations/Deployment-Guide_ZH.md.

Data Ingestion (Local)

Run ingestion from chatdku/chatdku/ingestion:

python update_data.py --data_dir ./data --user_id Chat_DKU -v True
python load_chroma.py --nodes_path ./data/nodes.json --collection_name chatdku_docs
python -m chatdku.chatdku.ingestion.load_redis --nodes_path ./data/nodes.json --index_name chatdku

Optional Django Stack (Docker profile)

docker compose --profile django up --build

This starts PostgreSQL, Django, and Celery in addition to the default services.

Repository Layout

• chatdku/: core agent logic, ingestion, backend, and frontend
• scraper/: recursive web scraper
• benchmarks/: benchmarking scripts
• Documentations/: internal docs (sanitized)

Notes

• No production data or secrets are included in this repository.
• Use .env or .env.local to provide your own credentials.