Sail is an AI infrastructure company focused on making high-quality open-source models more affordable and efficient to use, particularly for long-horizon agents, deep research, evaluations, and other sustained workloads. Its inference API offers pay-per-token pricing, monthly free credits, and optional completion windows that let customers trade response speed for lower costs. Developers can access a curated selection of models, including offerings from Kimi, DeepSeek, GLM, Gemma, and Qwen, through APIs compatible with OpenAI and Anthropic tools.

Beyond inference, Sail supports serving custom LoRA fine-tunes and running reinforcement-learning rollouts, with Tinker integration. Its Sailboxes are full virtual machines designed to let agents run for extended periods while retaining state and using compute to read, write, build, and test. Customers can start with a self-serve plan, choose Pro for expanded capacity and support, or contact Sail for enterprise services such as custom model deployments, volume pricing, and service-level agreements. Sail says it serves trillions of tokens weekly and provides zero data retention by default.