Poolside Debuts Laguna-XS.2 for Local Code Automation

By vramkickedin | | 2 min read
A compact smooth dark titanium cube floating gently just above a sleek silver laptop keyboard.

Laguna-XS.2 is an open source language model built specifically for automated programming tasks and extended project workflows. Built with a modular design, the system activates only three billion units of information at a time to maintain speed while handling complex requests.

Poolside developed the tool to run directly on personal computers, allowing creators to process sensitive code without relying on external servers. Local execution gives users full control over their data while maintaining strong performance across standard testing environments.

Model Size: 66.9GB & VRAM GPU: requirements vary

Agentic coding with local memory control

  • Mixes focused context review with standard memory tracking to speed up responses and save system resources.
  • Uses lightweight data compression for recent conversations to lower overall hardware demands.
  • Includes built-in reasoning steps that can be turned on or off for every request.
  • Runs smoothly on modern laptops equipped with thirty-six gigabytes of system memory.

Small engineering teams handling proprietary software can keep their entire workflow offline while still generating reliable automation scripts. Independent contractors working with limited hardware will appreciate the reduced memory footprint, which leaves plenty of room for other design and testing applications to run at the same time.

Architecture choices and benchmarking notes

Poolside trained the system using guided practice phases alongside automated task sorting to improve long-term project management. Testing scores show consistent results on verified coding challenges when measured against much larger competitors.

"Laguna XS.2 has native reasoning support and is designed to work best with preserved thinking, where reasoning content from prior assistant messages is preserved in the message history,"

noted the team on their model page. Developers can start using the framework through standard inference engines and terminal programs without navigating complex setup requirements. You can access the new Laguna-XS.2 model weights on Hugging Face.