Adaptive RAG app for building per-workspace knowledge bases from uploaded documents and chatting with them.

Includes:

• apps/api: FastAPI backend (auth, workspaces, uploads, chat/query)
• apps/web: Next.js web UI (workspaces, uploads, chat)
• src/ragify: core RAG pipeline (ingestion, context-aware chunking, embeddings/vector store, LangGraph routing)
• Infra: Docker Compose for MongoDB, Qdrant, and Grobid

• Storage: MongoDB for users/workspaces/sessions
• Vector DB: Qdrant per workspace collection
• Ingestion:
  • PDFs: Grobid-based parsing (batch) + indexing
  • Markdown/other files: convert to Markdown, chunk, index
• Generation: LangGraph workflow that routes queries (index vs general vs search) and can optionally call Tavily search

From repo root:

./run.sh start

This uses infra/docker/docker-compose.yml with env from infra/.env.

Copy env templates and adjust values:

cp .env.example .env
cp infra/.env.example infra/.env

Common variables (see .env.example):

• MONGODB_URI
• VECTORDB_URL (Qdrant)
• JWT_SECRET
• LLM_URL, LLM_MODEL, LLM_API_KEY
• TAVILY_API_KEY (optional, for web search route)

cd apps/api
./run.sh dev

API runs on http://localhost:8000 and exposes routes under /api/v1.

cd apps/web
pnpm install
pnpm dev

Default UI: http://localhost:3000.

• POST /api/v1/auth/register, POST /api/v1/auth/login
• GET /api/v1/workspaces, POST /api/v1/workspaces
• POST /api/v1/workspaces/{workspaceId}/upload (multi-file)
• POST /api/v1/workspaces/{workspaceId}/query
• GET /api/v1/workspaces/{workspaceId}/sessions

• apps/api/Dockerfile: builds and runs uvicorn apps.api.main:app
• apps/web/Dockerfile: builds Next.js standalone output
• infra/docker/docker-compose.yml: MongoDB, Qdrant, Grobid

• Workspace data is stored under apps/api/storage/workspaces/<workspace_id>/.
• Qdrant collections are created per workspace_id.