Turn PDFs, Word Docs, and Images into Instant Answers with WeKnora’s LLM‑Powered Search

Overview

WeKnora is an open‑source framework built on large language models (LLM) that enables semantic search and question‑answering over complex, multi‑format documents such as PDFs, Word files, and images.

Key Features

Agent mode : Supports ReACT‑style agents that can call built‑in tools, MCP tools, and web search, iterating and reflecting to produce comprehensive reports.

Precise understanding : Extracts structured content from PDFs, Word, images and creates a unified semantic view.

Intelligent reasoning : Leverages LLMs to comprehend document context and user intent for accurate QA and multi‑turn dialogue.

Multi‑type knowledge bases : Handles FAQ and document knowledge bases, with folder, URL import, tagging and online entry.

Flexible extensibility : Decoupled pipeline from parsing, embedding, retrieval to generation, allowing easy integration and customization.

Effective retrieval : Combines keyword, vector and knowledge‑graph strategies for cross‑knowledge‑base search.

Web search : Built‑in DuckDuckGo engine for extensible internet search.

MCP tool integration : Extends agent capabilities with uvx, npx launch tools and multiple transport methods.

Conversation strategy : Configurable agent model, normal model, retrieval thresholds and prompts to control multi‑turn behavior.

Simple UI : Intuitive web interface and standard API with zero technical barrier.

Secure and controllable : Supports on‑premise and private‑cloud deployment; data remains fully under user control.

Technical Architecture

The system is modular and separates the document understanding pipeline into five layers:

Document processing layer : Parses and preprocesses various formats (PDF, Word, images).

Knowledge modeling layer : Uses embedding, chunking, and knowledge‑graph techniques to build deep semantic representations.

Retrieval engine layer : Innovatively fuses keyword, vector and graph‑based retrieval for efficient and accurate recall.

Reasoning‑generation layer : Applies LLMs for deep understanding and answer generation, with integrated agent reasoning.

Interaction layer : Provides a web UI and standard REST API.

The design allows flexible combination of retrieval strategies, interchangeable LLMs (e.g., Ollama with Qwen, DeepSeek), and any vector database, while ensuring full controllability for private deployments.

Quick Start

Environment Requirements

Install the following tools locally:

Docker

Docker Compose

Git

Installation Steps

Clone the repository:

git clone https://github.com/Tencent/WeKnora.git
cd WeKnora

Copy the example environment file and edit .env with your configuration:

cp .env.example .env
# edit .env as needed

Start the services (includes Ollama):

./scripts/start_all.sh
# or
make start-all

Stop the services when finished:

./scripts/start_all.sh --stop
# or
make stop-all

Service Access

Web UI: http://localhost Backend API: http://localhost:8080 Jaeger tracing:

http://localhost:16686

Demo Screenshots

Knowledge base management, dialogue settings, and agent tool‑calling process are illustrated in the following images:

Open‑Source Repository

GitHub: https://github.com/Tencent/WeKnora