know3

Local RAG + Training Data Engine
Connecting to Ollama...

know3 v3.21

Local RAG + Training Data Engine · 100% Offline · by Dr. Khaled Diab
Connecting to Ollama... |
📚

Drop documents here to ingest

PDF · EPUB · DOCX · TXT — parsed, chunked, embedded & stored locally

0
Documents
0
Chunks
0
Training Pairs
0
Collections
👁️ Review Chunks (Optional - Filter before generating)

Review and filter chunks before generating pairs to improve quality and reduce processing time.

Process every Nth chunk (1 = all)
0 selected
Ingest documents first to see chunks
Tip: Select the chunks you want to generate pairs from. Use "Sample Mode" to test with every Nth chunk first.
Adjusts system prompt for domain
Processing... 0%

Generated Pairs

🔧 Tools & Advanced Features

📤 Load Custom Data

Upload JSON or JSONL files. Fields are auto-detected — works with any key structure.

📁
Drop JSON/JSONL files here
or click to browse · multiple files supported
Leave blank to auto-detect the longest text field

💬 RAG Conversation

Chat with your documents · Multi-turn · Source citations

🧠
Ingest documents first, then ask questions here
⚙️ Advanced Options
0.7
Step 1: Install Ollama → ollama.com/download
Step 2: Pull models:
  ollama pull llama3.2:3b (LLM)
  ollama pull nomic-embed-text (Embeddings for RAG)
Step 3: Set CORS (one time):
  • Mac: launchctl setenv OLLAMA_ORIGINS "*" then restart Ollama
  • Windows: Set env variable OLLAMA_ORIGINS=* in System Properties
  • Linux: OLLAMA_ORIGINS="*" ollama serve
Step 4: Refresh this page

⚙️ Settings

🌐 Ollama Connection
Checking...
Default: http://localhost:11434 · For LAN: http://192.168.x.x:11434
☁️ Cloud Inference (Faster)
Cloud = faster inference but documents leave your machine
🤖 Models
Language model for generation
Vector embedding for search
🔍 RAG Settings
How RAG queries work
Creativity (0.1=precise, 1.0=creative) 0.7
How many chunks to retrieve
Show sources in RAG responses
⚡ Generation
Content type (adjusts prompts)
Training pairs to generate
🎨 Theme
Emerald
Ocean
Sunset
Cyber Pink
Royal Gold
Monochrome
Pick any color — the full UI adapts automatically
🔧 Advanced
Text split size (200-3000)
Where new chunks are stored
Remove all chunks and pairs