Skip to content

configuration.md

Latest commit

 

History

History
64 lines (46 loc) · 2.1 KB

configuration.md

File metadata and controls

64 lines (46 loc) · 2.1 KB

Configuration Guide

This document explains the main configuration options available in ClawRAG.

Environment Variables

All configuration is done through the .env file. Here are the most important settings:

Basic Configuration

  • PORT: External port (default: 8080)
  • DOCS_DIR: Directory for document ingestion (default: ./data/docs)

LLM Configuration

  • LLM_PROVIDER: Choose between ollama, openai, anthropic, gemini, openai_compatible (default: ollama)
  • LLM_MODEL: The model to use (default: llama3:latest)

Embedding Configuration

  • EMBEDDING_PROVIDER: Usually matches LLM_PROVIDER (default: ollama)
  • EMBEDDING_MODEL: Embedding model (default: nomic-embed-text)

Ingestion Configuration

  • CHUNK_SIZE: Size of text chunks (default: 512)
  • CHUNK_OVERLAP: Overlap between chunks (default: 128)
  • INGEST_BATCH_SIZE: Batch size for ingestion (default: 10)

Application Settings

  • DEBUG: Enable debug logging (default: false)
  • LOG_LEVEL: Logging level (default: INFO)

Provider-Specific Configuration

Ollama (Default)

Uses the built-in Ollama container. No additional configuration needed.

OpenAI Compatible (LM Studio, Llama.cpp, etc.)

Set:

LLM_PROVIDER=openai_compatible
OPENAI_BASE_URL=http://host.docker.internal:1234/v1

OpenAI, Anthropic, Gemini

Set the appropriate API key in the docker-compose.yml file or environment.

Performance Tuning

Memory Management

  • The context window is limited to 8192 tokens to prevent OOM errors
  • For 8GB VRAM systems, use smaller models like llama3.2 (3b) or llama3 (8b)

Chunking Strategy

  • Smaller CHUNK_SIZE values provide more granular search results
  • Larger CHUNK_OVERLAP values help maintain context across chunks
  • Adjust based on your document types and search requirements

Troubleshooting

Common Issues

  • If ingestion fails, check that your DOCS_DIR is correctly mounted
  • If search returns no results, verify that embeddings are working properly
  • For LLM connectivity issues, ensure the provider is properly configured

Logging

Enable DEBUG=true to get more detailed logs for troubleshooting.