GitHub radar
Laguna XS 2.1: Poolside AI coding agent on Ollama
Poolside AI's coding-specialist model is now on Ollama — a 33B mixture-of-experts agent with 70.9% on SWE-bench Verified, designed to handle long software engineering tasks locally.
Laguna XS 2.1 is a coding-focused model from Poolside AI, a company that builds AI specifically for software engineering. The architecture is mixture-of-experts with 33B total parameters and only 3B active at inference time, which keeps memory usage manageable despite the scale. SWE-bench Verified score of 70.9% reflects solid real-world coding task performance. The model reasons natively between tool calls and supports a 262,144-token context window — enough to hold an entire project in one session. Poolside designed it specifically to run on a local machine. The Q4 quantized version weighs around 20 GB.
Why a vibe-coder should care
Long-horizon coding tasks — the kind where an agent needs to understand the whole repo and act across many files — previously required a cloud API. Laguna XS 2.1 brings that capability to a local Mac. Poolside's focus on software engineering means the model is tuned for exactly this use case, not a generic assistant that happens to write code.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install Laguna XS 2.1 from Poolside AI locally: ollama run laguna-xs-2.1 — then show me how to give it a full-project coding task
A Mac with 36 GB of RAM or more, or a GPU with 24 GB — the 4-bit version takes around 20 GB.
Open on Hugging Face▌ More finds