docrag-llm 0.1.16

Docling → Chroma → Ollama: simple, RAG pipeline

📄 DocRAG LLM

DocRAG LLM is a simple, Retrieval-Augmented Generation (RAG) pipeline.
It connects Docling (document parsing) → ChromaDB (vector store) → Ollama (local LLMs) into a single workflow, with both a CLI and a Python API.

---

✨ Features

• 🔍 Parse documents with Docling (PDF, DOCX, PPTX, HTML, etc.).
• 📑 Chunk text intelligently for retrieval.
• 🧠 Store embeddings in ChromaDB.
• 🤖 Answer questions using Ollama (default: llama3.2:1b).
• 🛡️ Designed for local execution (no cloud lock-in).
• 🖥️ Works both as a CLI tool and a Python library.

---

📦 Installation

pip install docrag-llm

Requirements

• Python 3.10+
• Ollama installed and running
• Models pulled locally:

  ollama pull llama3.2:1b
  ollama pull nomic-embed-text
  

---

🚀 Quickstart (CLI)

Ingest a document into Chroma

python -m docrag.cli ingest https://arxiv.org/pdf/2408.09869   --persist ./.chroma   --collection demo

• --persist → directory for Chroma DB (default: ./.chroma)
• --collection → logical collection name (default: demo)
• --embed → embedding model (default: nomic-embed-text)

---

Ask a question (default LLM = llama3.2:1b)

python -m docrag.cli ask "Give a concise bullet summary of the paper's main contributions."   --persist ./.chroma   --collection demo

• --llm → LLM model to use (default: llama3.2:1b)
• --top-k → number of chunks retrieved (default: 5)

---

Export parsed text

python -m docrag.cli export https://arxiv.org/pdf/2408.09869   --out-dir ./exports

Saves parsed text (Markdown/JSON).

---

CLI Help

python -m docrag.cli --help
python -m docrag.cli ingest --help
python -m docrag.cli ask --help
python -m docrag.cli export --help

---

🐍 Usage as a Python Library

from docrag import DocragSettings, RAGPipeline

# Configure pipeline
cfg = DocragSettings(
    persist_path="./.chroma",
    collection="demo",
    embed_model="nomic-embed-text",
    llm_model="llama3.2:1b",
)

pipeline = RAGPipeline(cfg)

# Ingest a document
n_chunks = pipeline.ingest("https://arxiv.org/pdf/2408.09869")
print(f"Ingested {n_chunks} chunks")

# Ask a question
answer = pipeline.ask("Give a concise bullet summary of the paper's main contributions.")
print(answer)

---

⚙️ Configuration

Both the Python API and CLI allow controlling:

• persist_path → path to Chroma DB
• collection → collection name
• embed_model → embedding model (Ollama tag)
• llm_model → LLM model (default: llama3.2:1b)
• chunk_chars / chunk_overlap → chunking granularity

---

📊 Roadmap

• Add model-check CLI command to list installed Ollama models.
• Support multiple backends (Weaviate, Milvus).
• Add streaming output for long answers.
• Expand test suite with large document regression cases.

---

🤝 Contributing

PRs and issues welcome! Please run lint and tests before submitting:

ruff check .
pytest

---

📜 License

MIT License © 2025