AtlasKB

A single-user, local-first knowledge base + RAG chat workbench.

Drop in your notes (Markdown / PDF / DOCX / PPTX / CSV / TXT), chat with an OpenAI assistant that cites the matching snippets, and have every Q&A turn auto-deduped and written back so retrieval gets sharper as you use it.

Local-only by design — the server binds to 127.0.0.1, no login layer.

Features

• Smart Markdown-aware chunker with sentence-level overlap and tiktoken budgeting
• SSE chat streaming with citation pills (KB / chunk position) and an "exclude prior Q&A from RAG" toggle
• Async, fail-safe Q&A write-back (chat UX is never blocked)
• Cross-KB BM25 search with <mark> highlights, KB filter sidebar, and Flat / Grouped views

Architecture

React + Vite UI  ──►  FastAPI (127.0.0.1:8000)
                          ├─ OpenAI (gpt-4o + ada-002)
                          └─ Elasticsearch 7.13.4 (Docker)

All persistent state lives in deploy/elasticsearch/data/.

Setup

# 1. Elasticsearch + Kibana
cd deploy && echo 'ELASTIC_PASSWORD=changeme' > .env && docker compose up -d && cd ..

# 2. Backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cat > .env <<EOF
OPENAI_API_KEY=sk-...
ELASTICSEARCH_URL=http://127.0.0.1:9200
ELASTIC_USERNAME=elastic
ELASTIC_PASSWORD=changeme
EOF
python main.py            # http://127.0.0.1:8000

# 3. Frontend
cd frontend && npm install && npm run dev   # http://127.0.0.1:5173

Swagger UI: http://127.0.0.1:8000/swagger-ui. Reset everything: stop containers and rm -rf deploy/elasticsearch/data.