Hey folks!
You know how AI coding assistants forget everything between sessions? You explain your codebase on Monday, and by Tuesday it's suggesting code that already exists.
That's the context window problem. Every AI has a limit on what it can hold in memory. Your codebase is bigger than that limit.
I built Cortex to fix this.
How It Works
Cortex indexes your codebase into a local vector database. When your AI needs context, it queries Cortex via MCP and gets the relevant code back.
It's smart about what to index - skips node_modules, build artifacts, and other noise. Only syncs on git commit, so your AI only sees verified code.
Everything runs locally. PostgreSQL + Ollama. No cloud, no API costs.
Quick Start
git submodule add https://github.com/Remskill/Cortex.git cortex
cd cortex && npm install
docker-compose up -d
npm run setup
5 minutes to set up. Check the README for MCP config.
Links
GitHub: github.com/Remskill/Cortex
MIT licensed. Free forever. Let me know what you think!