01-simple-rag

01 · Personal Research Assistant

The simplest CtxVault setup: one vault, one agent, local semantic search over a document collection.

This example is the entry point — no multi-agent topology, no persistent memory across sessions. Just a vault indexed over a set of documents and a LangChain RAG pipeline querying it semantically.

---

Scenario

A local collection of research documents — PDFs, markdown notes, plain text articles — indexed into a single vault and queried in natural language. Answers are grounded in the retrieved content, with citations back to the source documents.

---

What this demonstrates

• Initializing a vault and indexing a multi-format document collection
• Semantic retrieval as a LangChain retriever
• Grounded answer generation with source attribution
• Full local execution — no cloud dependencies

---

Setup

python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
export OPENAI_API_KEY=your_key
python app.py

Any LLM works — replace ChatOpenAI with Ollama or any LangChain-compatible provider.

---

Project structure

01-simple-rag/
├── .ctxvault/                        # Pinned vault config (commit this)
│   ├── config.json
│   └── vaults/
│       └── personal-vault/
│           ├── rag_survey_paper.pdf
│           ├── personal_notes.md
│           ├── blog_article.txt
│           └── rag_comparison.txt
├── app.py                            # RAG pipeline (~100 lines)
└── requirements.txt

---

Example output

QUERY: What are the main benefits of using RAG over fine-tuning?

Retrieved from:
  - rag_comparison.txt
  - personal_notes.md

ANSWER:
RAG avoids retraining entirely — knowledge updates require only adding
documents to the vault. Retrieved content provides direct citations,
and the data never leaves your infrastructure.

---

Next

Example 02 introduces multiple agents with isolated vaults and access control — the same vault primitive, composed into a multi-agent topology.