Artem Tyurin

Engineering @poolsideai. Previously at @apple and @uber.

[antirez @antirez]

Now with the upcoming release of the new DS4F checkpoint, and I believe later with Laguna S2.2 or a new S2.1 checkpoint, DwarfStar is going to be better at zero development costs. It is important to invest in the right models that get better and are locally runnable.

[Zed @zeddotdev]

Congrats to @poolsideai on launching Poolside Desktop Assistant.

Their Laguna models power an ACP agent that runs in any client, Zed included, and now their desktop app is an ACP client for any agent in the public registry.

Awesome to see Poolside taking full advantage of ACP!

[Artem Tyurin @tema_codes]

One user interface, many agents, zero lock-in.

[Poolside @poolsideai]

Today, we’re releasing Poolside Desktop Assistant.

One place to run coding agents across macOS, VS Code, and Visual Studio.

We built it for ourselves and have used it every day for the past year. Now we’re opening it up to everyone.

[antirez @antirez]

I just published mixed q2/q3 quants of Laguna S2.1 that work in MacBook systems with 64GB of RAM with quality not far from q4. You need DwarfStar laguna-s2.1 brach for them to work, or llama.cpp. With DwarfStar the speed is 65 t/s generation, ~570 t/s prefill.

[swyx @swyx]

one thing i think people dont appreciate enough about @poolsideai is their unusual degree of openness — not only have they shipped an excellent Small model that somehow beat @thinkymachines at coding, but most people (like @eliebakouch) have been shouting out their excellent papers, but also they're among a rare few to actually expose their full eval dataset as well - beautifully published, with across 6 public benchmarks with 4 runs each and hundreds of turns per run. you can satisfy for yourself if they rewardhack. brilliant.

[Artem Tyurin @tema_codes]

@saik0s cooking by the pool

[Artem Tyurin @tema_codes]

Open models and open protocols.

Try Laguna S 2.1 in pool, our coding agent.

Works in the terminal and in any Agent Client Protocol client, like Zed, JetBrains, Xcode, and Neovim.

[Poolside @poolsideai]

Today we're releasing Laguna S 2.1, our most capable model to date.

It's a 118B total parameter Mixture-of-Experts model with 8B activated per token, a context window of up to 1M tokens, and thinking and no-thinking modes.

Capable enough to hold its own against models many times its size. Small enough to run on a single @NVIDIAAI DGX Spark.

Laguna S 2.1 is fully open under OpenMDW-1.1, with weights available today on @huggingface

https://t.co/xxGeAgo35R

[NVIDIA AI @NVIDIAAI]

Congrats to the @poolsideai team on their latest release, Laguna S 2.1.

It’s open-weight, delivers way beyond its size, runs great locally and you can customize it with NVIDIA NeMo.

Go try it out on @OpenRouter or download from @huggingface 👉https://t.co/WOIvTVC7wO

[Artem Tyurin @tema_codes]

https://t.co/AVMtmH6tA9

[Pengming Wang @PengmingWang]

Quite excited about this one: Laguna S 2.1: 118B-A8B MoE with 1M context. It runs on a single NVIDIA DGX Spark, and it stacks up very well on what we focus on: agentic coding.

A few observations we had building it. 🧵

[Poolside @poolsideai]

Turns out, the next Laguna also fits on a @NVIDIAAI DGX Spark 👀

[elie @eliebakouch]

new jspace results :) this time focusing on @poolsideai laguna XS.2 and XS.2.1 version bump and optimizers

for laguna, the question i tried to answer is "XS.2 -> XS.2.1 is likely training the model on more/different code, so how different is the jlens between the two versions if i use code or wikitext for the prompt?"

and the answer is that it's indeed VERY different. the model workspace is totally different between XS.2 and XS.2.1 if i fit it on code (i use some of the agentic traces training data from nemotron), and quite similar if i do it on wikitext EXCEPT the end layers that are similar for both (called the "motor" section in anthropic paper)!

confounding factor is that i took longer prompts for code compared to wikitext, fable and kimi K3 said it was not very important tho (i might test it to be sure!)

for the optimizer (based on checkpoints from the fantastic optimizer paper by @wen_kaiyue and marin folks), the question was a bit more open ended. we found that the CKA between different optimizers is much lower (less similar) than between the same optimizer at different sizes. we also found that some optimizers are much more similar than others and that the PR (effective number of directions a layer uses to talk to the logits) is quite different (see other plots in thread!)

the overall structure is still relatively similar, feels like muon has fewer "outlier" layers, the ones that have lower CKA with other layers and create those lines in the heatmap

[Artem Tyurin @tema_codes]

Always great to see more ACP clients. At @poolsideai, we have pool TUI and something coming up soon for the GUI fans as well, stay tuned!

[JetBrains Air @getsome_air]

Air now supports 19 agents:
Amp, Augment Code, Cline, Codex CLI, Cursor, Devin, Factory Droid, GitHub Copilot, Hermes Agent, Junie, Mistral Vibe, Pi, Stakpak, Autohand Code, Claude Agent, Kilo, Kimi CLI, OpenCode, and Poolside.

2/5

[JetBrains Air @getsome_air]

Air now supports 19 agents:
Amp, Augment Code, Cline, Codex CLI, Cursor, Devin, Factory Droid, GitHub Copilot, Hermes Agent, Junie, Mistral Vibe, Pi, Stakpak, Autohand Code, Claude Agent, Kilo, Kimi CLI, OpenCode, and Poolside.

2/5

[Artem Tyurin @tema_codes]

Try it out https://t.co/AVMtmH6tA9

[Artem Tyurin @tema_codes]

Open protocols in action: pool connecting to a Goose agent on an exe.​dev VM via the new ACP network transport.

[Poolside @poolsideai]

Today we’re releasing Laguna XS 2.1.

It’s a small upgrade to the Laguna XS.2 model, the same 33B total / 3B active MoE and stronger results on multilingual coding and terminal-style tasks.

Available now on @huggingface, @OpenRouter, and via Poolside API.

[Artem Tyurin @tema_codes]

@ntaylormullen Can we get ACP support for Antigravity?

[Poolside @poolsideai]

IYMI: the best way to try Laguna M.1 is to jump in the pool.

pool is our agent harness. It works as both an ACP server and client, so you can run M.1 as a coding agent and build with the same interface we use ourselves.

go build something cool ↓
https://t.co/AgzWth83Ll