keovil 0.1.0

Private RAG System for Documents & Structured Data

Keovil

Private Query Interface for Documents & Structured Data

Ask questions in plain English. Keovil queries your files.

Upload PDFs, text files, code, and more. Or query CSV, Excel, SQLite databases via the web app. Keovil generates the queries and returns results. Document processing runs locally on your GPU, with flexible LLM options (cloud or local).

Installation

Option 1: Install from GitHub (Recommended)

# Install the SDK and web app
pip install git+https://github.com/kaiserkonok/Keovil.git

Option 2: Local Development Install

# Clone the repository
git clone https://github.com/kaiserkonok/Keovil.git
cd Keovil

# Install in development mode
pip install -e .

Option 3: Web App Only (Quick Start - No Docker!)

# 1. Install Keovil
pip install git+https://github.com/kaiserkonok/Keovil.git

# 2. Run the web app (Qdrant runs automatically!)
python -m keovil_web

That's it! Qdrant runs in embedded mode automatically.

---

Optional: Run Qdrant Manually (For Better Performance)

If you want to run Qdrant externally (e.g., via Docker) for better performance:

# macOS
brew install qdrant && brew services start qdrant

# Linux (Docker)
docker run -d -p 6333:6333 -v qdrant_storage:/qdrant_storage qdrant/qdrant

Then set QDRANT_HOST=your-server if not on localhost.

---

Quick Start

Web App

# 1. Install Keovil
pip install git+https://github.com/kaiserkonok/Keovil.git

# 2. Run the web app (Qdrant runs automatically!)
python -m keovil_web

Open http://localhost:5000

That's it! Qdrant runs in embedded mode automatically. No Docker needed.

For local LLM, also install Ollama: curl -fsSL https://ollama.com/install.sh | sh

SDK (For Developers)

from keovil import KeovilRAG
from keovil.utils.llm_config import LLMConfig

# Initialize with default Ollama
rag = KeovilRAG(data_dir="/path/to/your/files")

# Or use a specific LLM provider
config = LLMConfig(provider="openai", model="gpt-4o", openai_api_key="sk-...")
rag = KeovilRAG(data_dir="/path/to/files", llm_config=config)

# Index your files (PDF, text, code, etc.)
rag.ingest(["document1.pdf", "notes.txt"])

# Ask questions in natural language
answer = rag.query("What is the recommended dosage for adults over 65?")
print(answer)

# For chat-like conversations
history = []
answer = rag.query("What are the API rate limits for the Pro plan?", history)
history.extend([("You", "What are the API rate limits for the Pro plan?"), ("AI", answer)])
answer = rag.query("Does the Enterprise plan include SSO?", history)

---

Linux Installation

# Option 1: Just Keovil (Qdrant auto-runs in embedded mode)
pip install git+https://github.com/kaiserkonok/Keovil.git
python -m keovil_web

# Option 2: With local Ollama (if using local LLM)
curl -fsSL https://ollama.com/install.sh | sh
ollama pull qwen2.5-coder:7b-instruct
pip install git+https://github.com/kaiserkonok/Keovil.git
python -m keovil_web

# Option 3: With external Qdrant (for better performance)
pip install git+https://github.com/kaiserkonok/Keovil.git
curl -L https://github.com/qdrant/qdrant/releases/download/v1.7.4/qdrant-linux-amd64.tar.gz -o qdrant.tar.gz
tar -xzf qdrant.tar.gz
./qdrant &
python -m keovil_web

---

Features

🗄️ Structured Data Analysis (Web App)	Query CSV, Excel, SQLite, Parquet via natural language. Keovil generates and executes SQL via DuckDB.
📄 Document Q&A	Ask questions about PDFs, text files, code. Built on ColBERT retrieval with Qdrant. Docling parses documents locally.
🔄 Automatic Indexing	Drop files in a folder — Keovil syncs and indexes them automatically.
🔒 Total Privacy	Document processing runs locally. Cloud LLMs are optional - use local Ollama for full privacy.
🌐 Multi-LLM Support	Use Ollama (local), OpenAI, Anthropic, OpenRouter, or Gemini. Change anytime without restart.
⚡ Flexible Hardware	Full GPU not required - can use cloud LLMs with local document processing.

---

Supported LLM Providers

Web App

Change providers anytime via Settings page - no restart needed!

Provider	Description	API Key Required
Ollama	Local models running on your machine	No
OpenAI	GPT-4o, GPT-4o-mini, etc.	Yes
Anthropic	Claude 3.5 Sonnet, Haiku, etc.	Yes
OpenRouter	Access 100+ models via single API	Yes
Gemini	Google Gemini 2.0, 1.5 Pro, etc.	Yes

SDK (For Developers)

from keovil import KeovilRAG
from keovil.utils.llm_config import LLMConfig

# Ollama (default)
rag = KeovilRAG(data_dir="/path", llm_config=LLMConfig(provider="ollama", model="qwen2.5-coder:7b"))

# OpenAI
config = LLMConfig(provider="openai", model="gpt-4o", openai_api_key="sk-...")
rag = KeovilRAG(data_dir="/path", llm_config=config)

# Anthropic
config = LLMConfig(provider="anthropic", model="claude-3-5-sonnet-20241022", anthropic_api_key="sk-ant-...")
rag = KeovilRAG(data_dir="/path", llm_config=config)

# OpenRouter
config = LLMConfig(provider="openrouter", model="openai/gpt-4o-mini", openrouter_api_key="sk-or-...")
rag = KeovilRAG(data_dir="/path", llm_config=config)

# Gemini
config = LLMConfig(provider="gemini", model="gemini-2.0-flash", gemini_api_key="AIza...")
rag = KeovilRAG(data_dir="/path", llm_config=config)

---

Architecture

┌─────────────────────────────────────────────────────────────────────────────┐
│                              KEOVIL INTERFACE                                │
│                    Flask Web UI (localhost:5000)                            │
│                    └── Settings: Switch LLM providers                       │
└───────────────────────────────┬─────────────────────────────────────────────┘
                                │
              ┌─────────────────┼─────────────────┐
              ▼                 ▼                 ▼
┌─────────────────────┐ ┌─────────────────────┐ ┌─────────────────────┐
│    KNOWLEDGE LAB    │ │   STRUCTURED DATA   │ │   CMS EXPLORER      │
│   (Document Q&A)    │ │       LAB           │ │   (File Manager)    │
│                     │ │   (SQL Queries)      │ │                     │
└──────────┬──────────┘ └──────────┬──────────┘ └──────────┬──────────┘
           │                        │                        │
           ▼                        ▼                        │
┌─────────────────────┐ ┌─────────────────────┐            │
│    COLBERT ENGINE   │ │   SQL QUERY AGENT   │            │
│   (Qdrant + Torch)  │ │     (DuckDB)        │            │
└──────────┬──────────┘ └──────────┬──────────┘            │
           │                        │                        │
           ▼                        ▼                        │
┌─────────────────────┐ ┌─────────────────────┐            │
│       QDRANT       │ │       DUCKDB        │            │
│  Vector Database   │ │  SQL Engine (GPU)   │            │
└─────────────────────┘ └─────────────────────┘            │
           │                        │                        │
           ▼                        ▼                        │
┌─────────────────────────────────────────────────────────────┐
│                        LLM PROVIDERS                         │
│   ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌────────┐ │
│   │ Ollama  │ │ OpenAI  │ │Anthropic│ │OpenRouter│ │ Gemini │ │
│   │ (Local) │ │  GPT-4  │ │Claude-3 │ │ 100+ LMs │ │Gemini  │ │
│   └─────────┘ └─────────┘ └─────────┘ └─────────┘ └────────┘ │
│              Dynamic switching - no restart needed!          │
└─────────────────────────────────────────────────────────────┘

---

Data Flow

DOCUMENTS                          STRUCTURED DATA (Web App Only)
─────────                          ──────────────────────────────

User drops files ────────────────── User drops files
     │                                    │
     ▼                                    ▼
DocumentProcessor ────────────────── FileWatcher
(Extract text)                     (Monitor folder)
     │                                    │
     ▼                                    ▼
IntelligentChunker                  Auto-Sync to
(Smart splitting)                   DuckDB Views
     │                                    │
     ▼                                    ▼
ColBERT Embeddings                  SQL Tables
(Torch GPU)                         (CSV/XLSX/etc)
     │                                    │
     ▼                                    ▼
Qdrant VectorDB ────────────────── DuckDB
(Index & store)                    (Execute SQL)
     │                                    │
     │                                    │
     └──────────────┬───────────────────┘
                    │
                    ▼
              ┌───────────┐
              │   LLM    │
              │(Any Prov)│
              └─────┬─────┘
                    │
                    ▼
              User Answer

---

Requirements

Hardware

With Cloud LLM (OpenAI, Anthropic, Gemini, OpenRouter)

Component	Minimum	Recommended
GPU VRAM	6GB (RTX 3060)	8GB+ (RTX 4060/4070)
RAM	16GB	32GB

Cloud LLMs don't use local GPU - GPU only needed for Docling (document parsing) + ColBERT (embeddings). 6GB minimum tested and working; 8GB recommended.

With Local LLM (Ollama)

Component	Minimum	Recommended
GPU VRAM	10GB+	16GB+
RAM	16GB	32GB

Local Ollama + ColBERT + Docling all need GPU. 10GB minimum for 7B models + embeddings. 16GB recommended for smooth operation.

No GPU (CPU Only)

Component	Minimum
RAM	16GB

Works but extremely slow. Document ingestion minutes vs seconds. Not recommended for production use.

Software

Dependency	Version	Purpose
Python	3.12+	Runtime
CUDA	12.4+ (12.8 for RTX 50)	GPU acceleration (if using GPU)
Ollama	Latest	Local LLM (only if using local)
Qdrant	Optional	Vector database (auto-embedded if not running)

No Docker needed! Qdrant runs automatically in embedded mode if no external server is available. Everything works out of the box.

No GPU? Works with cloud LLMs only. Document processing will be slower but functional.

---

Configuration

Environment Variables

Variable	Default	Description
KEOVIL_PROVIDER	ollama	LLM provider (ollama, openai, anthropic, openrouter, gemini)
KEOVIL_MODEL	provider-specific	Model name
OLLAMA_HOST	127.0.0.1:11434	Ollama server address
OPENAI_API_KEY	-	OpenAI API key
ANTHROPIC_API_KEY	-	Anthropic API key
OPENROUTER_API_KEY	-	OpenRouter API key
GEMINI_API_KEY	-	Google Gemini API key
QDRANT_HOST	localhost	Qdrant server address
STORAGE_BASE	~/.keovil	Custom storage path

Storage Locations

~/.keovil/                 ← Default storage for both SDK and web app
├── data/                  # Source files
├── database/              # SQLite manifest + chat history
├── config.json            # LLM provider settings
└── qdrant/               # Vector embeddings

---

Usage Modes

Web Application

# After pip install, run:
python -m keovil_web

• Uses ~/.keovil for storage
• Collection: keovil_app
• Visit http://localhost:5000
• Go to Settings to change LLM provider anytime

SDK (For Developers)

from keovil import KeovilRAG

rag = KeovilRAG(data_dir="/path/to/files")
rag.ingest(["file.pdf"])
answer = rag.query("Your question?")

• Uses ~/.keovil for storage
• Collection: keovil
• Pass llm_config to use different providers

Custom Storage

# Override storage location
export STORAGE_BASE=/path/to/custom/storage
python -m keovil_web

---

Troubleshooting

Ollama Issues

# Check installation
ollama --version

# Check running
ollama serve

# Pull model
ollama pull qwen2.5-coder:7b-instruct

# List models
ollama list

Qdrant Issues

# Check health
curl http://localhost:6333/healthz

# Response should be: {"status":"ok"}

GPU Issues

# Check NVIDIA driver
nvidia-smi

# Check PyTorch CUDA
python -c "import torch; print(f'CUDA: {torch.cuda.is_available()}')"

# Check VRAM
nvidia-smi --query-gpu=memory.used,memory.total --format=csv

---

License

GNU General Public License v3.0 — see LICENSE for details.

---

Star · Issues · Website