The .context/ method

A new way to get AI to do what you want without expensive RAG solutions

Remember when I wrote about Documentation as Code as Context? Theory is nice. But you want to see it work.

So I built it.

The Problem Nobody Talks About

Everyone's building RAG pipelines. Vector databases, embedding models, chunking strategies, retrieval algorithms. The infrastructure alone costs more than most side projects make in a year.

But here's what they won't tell you:

for 90% of use cases, you're over engineering it.

You don't need to vectorise your entire codebase. You don't need semantic search over documentation. You just need AI to understand what you're building.

The Stupidly Simple Solution

Create a .context/ folder. Write markdown files. Feed them to AI.

That's it.

.context/
├── project.md       # What you're building
├── architecture.md  # How it's structured
├── methods.md       # Core patterns and approaches
└── rules.md         # Constraints and conventions

No vectors. No embeddings. No retrieval. Just text files that travel with your code.

Show Me The Code

I've open-sourced a working example: github.com/andrefigueira/.context

Here's what's inside:

project.md - Dead simple project overview:

markdown

# Project: AI-Powered Analytics Dashboard

A real-time analytics platform that uses AI to surface insights from user behavior data.

Tech stack: Next.js, PostgreSQL, OpenAI API

methods.md - Your patterns, your way:

markdown

# Authentication Method

We use JWT tokens with refresh rotation. No sessions. No cookies for auth.

# Data Processing Method  

Raw events → Kafka → Processing pipeline → PostgreSQL
Never process synchronously. Always queue.

architecture.md - The map of your system:

markdown

# System Architecture

- /api - Next.js API routes
- /lib - Shared utilities  
- /components - React components
- /workers - Background job processors

Database: Single PostgreSQL instance with read replicas
Caching: Redis for hot data only

The Workflow That Actually Works

1. Start a new feature? Update .context/methods.md with your approach

2. Change architecture? Document it in .context/architecture.md

3. Open your AI assistant? It reads these files first

No separate documentation site. No wiki that goes stale. Your context lives with your code, changes with your code, ships with your code.

But Does It Scale?

Fair question. Here's what I've learned:

It works great for:

• My projects

• Small teams who've tried it

• Focused, single-domain applications

• Rapid prototyping

You might need RAG when:

• You need semantic search across thousands of documents

• Multiple systems with conflicting conventions

• You literally can't summarise the important parts

But be honest, are you there yet?

The Money Shot

RAG pipeline setup:

• Vector database: $200-500/month

• Embedding API costs: $50-200/month

• Infrastructure maintenance: 10 hours/month

• Initial setup: 40-80 hours

Total: $250-700/month + setup time

Context files:

• Cost: $0

• Setup: 1 hour

• Maintenance: Part of normal development

Total: Free

Copy This Right Now

1. Clone the repo:

git clone https://github.com/andrefigueira/.context

2. Copy the .context/ folder to your project

3. Fill in the templates with your project details

4. Open Claude/GPT/Whatever and paste:

Here's my project context:

[paste your .context files]

Now help me build [whatever you're building]

The Uncomfortable Truth

Most developers won't do this. They'll keep complaining about AI not understanding their codebase while refusing to write three markdown files.

They'll spend weeks building embedding pipelines instead of spending an hour writing clear context.

Don't be most developers.

What's Next

This is version one. The repo is public. Fork it. Make it better.

Share what works. Share what doesn't.

Because the best solution isn't the most sophisticated one, it's the one you'll actually use.