Contextual

Contextual is a simple, general-purpose Retrieval-Augmented Generation (RAG) system for Markdown-based documentation. It parses, chunks, and embeds .md or .mdx files, stores them in a PostgreSQL vector database (pgvector), and serves scoped, context-aware answers via OpenAI's API (Embedding is done using text-embedding-3-small and prompts are sent to gpt-4o by default).

React documentation is used as the default dataset, but any Markdown content can be used with minimal configuration. To change the context source, just replace files in the data/ directory.

---

Getting Started

1. Clone the repository

git clone git@github.com:mykelkyle/contextual.git
cd contextual

2. Install dependencies

npm install

3. Build the project

npm run build

4. Create the PostgreSQL database

psql postgres
CREATE DATABASE contextual;

5. Run DB setup

node dist/setup.js

6. Generate and store embeddings

node dist/loadEmbeddings.js

7. Start chat interface

node dist/index.js

Environment Variables

Create a .env file in the root of the project with the following:

OPENAI_API_KEY=<your_openai_key>
DATABASE_URL=<your_db_url>

Future Work

• Web UI for chat and data upload
• Abstract away DB setup
• Support for additional file formats (.pdf, .html, etc.)