Poolside is emerging as a key player in open, secure AI and developer-focused agent tooling, as highlighted by a series of product and partnership announcements this week. The company joined NVIDIA’s Open Secure AI Alliance and expanded its Laguna S 2.1 model stack with a new multi-agent desktop workspace for coders.
Poolside’s participation in the NVIDIA-led alliance underscores its commitment to transparency and security in frontier AI systems. The company plans to contribute models, evaluations, and research openly, aiming to give cybersecurity defenders broader access to advanced AI intelligence.
This alignment with a security-focused consortium may reinforce Poolside’s credibility with enterprise and public-sector buyers. It also positions the firm for deeper collaboration opportunities in secure AI solutions, a segment where regulatory scrutiny and demand for trustworthy tooling are rising.
On the product front, Poolside introduced Desktop Assistant, a macOS application that orchestrates multiple coding agents across projects and repositories. The tool integrates with VS Code and Visual Studio and supports agents such as pool, Claude Code, Codex, and Gemini operating in parallel.
By building on ACP, an open protocol from Zed, and separating interface, agent harness, and model, Desktop Assistant is designed for interoperability and reduced vendor lock-in. The product reflects a shift from simple chat workflows toward supervising long-running, multi-repository agent activity.
Desktop Assistant ships with Laguna S 2.1 using Poolside’s internal agent reinforcement learning harness, and offers offline use via Laguna XS 2.1 through MLX. The company has relied on the assistant internally for about a year, indicating a degree of maturity and fit for professional developer use cases.
Laguna S 2.1 itself is an open-weight large language model engineered for agentic coding and long-horizon tasks. The 118-billion-parameter Mixture-of-Experts design uses 8 billion active parameters per token and supports up to 1 million tokens of context, with “thinking” and “no-thinking” modes to balance cost and performance.
The model targets persistent coding and research workflows, maintaining goals, using tools, and self-checking outputs over extended periods. This long-running autonomy is aimed at enterprise automation and software development where reliability, data control, and cost efficiency are central priorities.
Laguna S 2.1 is positioned as a self-hostable alternative to metered closed APIs, matching or outperforming larger systems on benchmarks like Terminal-Bench 2.1 and SWE-Bench Pro. Its compact footprint enables deployment on a single NVIDIA DGX Spark, supporting data sovereignty for regulated customers.
Trained via Poolside’s Model Factory platform, Laguna S 2.1 reflects an efficiency-first training strategy that uses automated architecture experimentation and reinforcement learning from code execution. The pipeline supports roughly five-week iteration cycles and training completion in under four weeks on 4,000 H200 GPUs.
The model is released under the OpenMDW-1.1 license with multiple checkpoints and quantizations, and is hosted on Hugging Face with broad compatibility across serving frameworks and the Poolside API. Strategically, Poolside is targeting Western enterprises and public-sector entities seeking trusted open-weight coding intelligence.
Taken together, the alliance membership, Desktop Assistant launch, and Laguna S 2.1 release mark a pivotal week for Poolside. The company is deepening its role in secure, open AI while pushing forward an integrated stack for enterprise-grade coding agents and AI-native development environments.