Poolside AI

About

Poolside attacks coding AI from the model layer up: it trains open-weight foundation models on software-producing data and ships the runtime around them, betting that code generation is the fastest route to generally capable systems. Its Laguna family spans Laguna S 2.1, a frontier-class 118B-parameter model (8B active) with 1M-token context, and Laguna XS 2.1, a 33B model (3B active) designed to run on-device with 256K context. Products include a desktop app with projects, connectors, and PR/branch context, a CLI, and an agent runtime in research preview, with the models also reachable through OpenRouter and Vercel AI Gateway. The company describes its training pipeline as a Model Factory that continuously evaluates and refines models against execution feedback. Its audience spans developers wanting an integrated agentic experience and teams integrating the open-weight models through gateways, with docs at docs.poolside.ai.