Cursor breaks when your codebase gets big. Here’s how to vibe code...

Here’s how to vibe code large projects ↓
Plandex is an AI-powered command-line tool designed for massive codebases.
It works like Cursor, but without choking on size.
Built to handle:
- 2 million tokens in context
- 20+ million tokens when indexing directories
Here’s how it works:
- It uses Tree Sitter project maps to understand code structure
- It auto-selects the best LLM (ChatGPT, Claude, Mistral, OpenAI, etc.) via OpenRouter
- It supports brainstorming + implementation through chat and tell modes
- All changes are sandboxed until you approve them
- Plandex Cloud (no API keys needed)
- Plandex Cloud with your own keys
- Self-hosted local mode (the one I used)
Here’s how to set it up locally:
Install Docker: docs.docker.com/get-docker/
Clone the repo:
git clone github.com/plandex-ai/pla…
Start the local server using the command from their GitHub
Docs: docs.plandex.ai/hosting/
- Open a new terminal → install the CLI (uses sudo)
- Sign in locally (it creates a user for you)
- When prompted, select local mode and confirm the host address (just press Enter)
- Set your OpenRouter and OpenAI API keys as environment variables
Navigate to your project directory, run: plandex init
This sets up your project so you can start coding with AI support.
Use:
- Chat mode to brainstorm features
- Tell mode to generate step-by-step code
- Auto mode to debug issues
Tried out Plandex and it looked super cool for big projects.
Haven’t used it on a real client build yet so can’t say much long-term.
Just wanted to share it in case it helps someone. Credits to the Plandex team for the demo.
Let me know if you try it out.