LangChain-Qdrant Multi-Source API

Developed by NexuSecurus™

Supported HW	Target OS	Code Stats	Audience	Type	Dependencies	License

Table of Contents

• LangChain-Qdrant Multi-Source API
• Table of Contents
  • Description
  • Features
  • Program Requirements / Prerequisites
    • Hardware
    • Software
  • Installation
    • Clone the repository
    • Environment configuration
    • Docker Compose (RECOMMENDED)
    • Python (Run Locally)
  • API Endpoints
    • > For a full detailed description of all API endpoints, request/response examples, default values, and n8n integration examples, please check ENDPOINTS.md.
    • LLM Generation
    • Ingestion Endpoints
    • Semantic Search
    • Collections
    • Debug Endpoints
    • Health
  • Contributing
  • License
  • References

Description

Langdrant is an open-source API backend for semantic document search, powered by Qdrant and modern embedding models. Easily ingest, embed, and search your documents with a simple REST API. Perfect for building AI-powered search, knowledge bases, and more. A high-performance FastAPI server for multi-source data ingestion, semantic search, and LLM-powered query generation. It integrates Qdrant as the vector database and Ollama for embeddings and LLM completions.

Supports asynchronous ingestion, streaming responses, and structured n8n-ready outputs.

Features

• Multi-source ingestion
  • Generic text
  • Files: PDF, DOCX, HTML, TXT
  • Logs
  • Database rows (PostgreSQL)
  • RSS feeds
  • Social media posts
• Vector search
  • Semantic search with optional LLM-based context
  • Hybrid queries (vector similarity + keyword filters)
  • Multi-collection search
• LLM generation
  • Context-aware completions
  • n8n-ready structured responses (summary and canonical_embedding_text)
  • Streaming responses via SSE for conversational endpoints
• Collection management
  • List collections with vector counts
  • Delete collections safely
• Debug endpoints
  • Text chunk preview
  • Embedding inspection
• Security
  • API key enforcement for all endpoints
• Deployment-ready
  • Dockerized with FastAPI & Uvicorn
  • Configurable via .env
  • Multi-worker async ingestion

---

Program Requirements / Prerequisites

Hardware

• CPU: x86_64 or ARM architecture
• RAM: Minimum 4GB (8GB recommended for optimal performance)
• Disk: Minimum 10GB free space
• Internet connection: Required for fetching dependencies, accessing APIs, and pulling container images
• Ollama Instance with available models (Local or Remote)

Software

• Git: Required to clone the repository and manage version control

  sudo apt install git  # Linux
  brew install git      # macOS

• Python 3.11+: Required for running locally or for building Docker images

  python3 --version

• pip: Python package manager to install dependencies

• Docker: Required for containerized deployment

  docker --version

• Docker Compose: For multi-container orchestration

  docker compose version

• PostgreSQL client libraries: Required if you intend to use database ingestion endpoints (psycopg2-binary included in Python dependencies)

• Optional system packages for file ingestion (PDF/DOCX/HTML parsing, OCR):

  • poppler-utils (PDF text extraction)
  • libreoffice (DOCX conversion, optional)
  • tesseract-ocr (OCR for scanned documents)

• cURL or HTTP client for testing API endpoints

• jq (optional) make json output prettier and more readable.

• n8n (optional) for automation workflows and connecting ingestion + query endpoints

---

Installation

Clone the repository

git clone https://github.com/nexusecurus/langdrant.git
cd langdrant

Environment configuration

Copy the .env.example to .env and configure your environment:

cp .env.example .env

Auto Generate API Key (Optional):

python3 langserver/api-generator.py

This will generate an API_KEY and update it under .env file automatically.

Customize variables to your preference:

nano .env

Variable	Description	Default
API_KEY	API key for FastAPI endpoints	""
API_PORT	FastAPI port	8000
LOG_LEVEL	Logging level	INFO
QDRANT_URL	Qdrant server URL	http://127.0.0.1:6333
QDRANT_API_KEY	Optional Qdrant API key	""
VECTOR_SIZE	Embedding vector size	1536
OLLAMA_BASE_URL	Ollama server URL	http://127.0.0.1:11434
EMBED_MODEL	Ollama embedding model	nomic-embed-text
LLM_MODEL	Ollama LLM model	llama3:8b
LLM_CTX	LLM context window	4096
LLM_MAX_TOKENS	Max tokens per generation	300
DB_HOST/PORT/NAME/USER/PASSWORD	PostgreSQL connection	-
CHUNK_SIZE	Text chunk size	800
CHUNK_OVERLAP	Overlap per chunk	120
EMBED_BATCH_SIZE	Batch size for embedding	64

Docker Compose (RECOMMENDED)

docker compose build --no-cache
docker compose pull
docker compose up -d

Python (Run Locally)

python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
cd langserver
uvicorn main:app --host 0.0.0.0 --port 8000 --workers 4

---

API Endpoints

> For a full detailed description of all API endpoints, request/response examples, default values, and n8n integration examples, please check ENDPOINTS.md.

LLM Generation

POST /generate
POST /chat

• Generate text using an LLM model. Returns structured n8n-ready data (summary and canonical_embedding_text). Use model, max_tokens, and num_ctx to override defau
• Conversational endpoint. Accepts a list of messages. Supports streaming mode via SSE. Returns assistant response. Flags: model, max_tokens, num_ctx, stream.

Ingestion Endpoints

Async ingestion of multiple data sources:

POST /ingest_texts       # Generic text
POST /ingest_file        # File upload
POST /ingest_logs        # Logs
POST /ingest_db          # Database rows
POST /ingest_rss         # RSS feeds
POST /ingest_social      # Social posts
POST /fetch_rss_feeds    # Fetch and ingest RSS feeds

• Supports deterministic ID generation for deduplication
• Text is chunked with configurable size and overlap
• Batch embedding and upsert into Qdrant
• Preserves full metadata (source, timestamp, platform, etc.)

Semantic Search

POST /query              # Single collection
POST /query_hybrid       # Hybrid semantic + keyword filters
POST /query_multi        # Multi-collection search

• Returns top-K nearest vectors
• Optional LLM answer generation from retrieved context
• Keyword filters, recency boosts, and hybrid queries supported

Collections

GET /collections
POST /collections/delete

• List collections with vector counts
• Delete collection safely

Debug Endpoints

POST /debug/chunk
POST /debug/embeds

• Preview how text is chunked
• Inspect embeddings before insertion

Health

GET /health
GET /ping

• Returns server status.

---

Contributing

1. Fork the repository
2. Create your feature branch
3. Submit pull requests with clear descriptions
4. Ensure all new endpoints have tests and documentation

---

License

MIT License

---

References

• FastAPI
• Qdrant
• LangChain
• Ollama API