Coasty

Coasty is a computer-use AI platform that combines a public arena, real desktop environments, evaluation infrastructure, and a computer-use API. Its site positions the company around the full stack of computer use: it runs Coarena, builds Windows and Linux environments for evaluation, trains production computer-use models, and offers public and private benchmarking. The developer docs frame the API as an entire computer exposed through one API, where a user can hand over a goal or take control of the plan, machine, or agent loop. For builders, the most relevant piece is the Coasty Computer Use API. The docs expose a task workflow through endpoints such as `POST /v1/tasks`, where callers can give Coasty the outcome they want completed. The model can own the goal, plan, machine, and loop, but the design also lets a developer take control of lower layers without changing key concepts. That makes Coasty useful for teams experimenting with computer-use agents that need real desktops rather than browser-only automation. Coasty also emphasizes evaluation quality. The homepage argues that stale public benchmarks, cherry-picked demos, and toy VMs do not prove real capability. Its stack uses real operating systems, real applications, outcome grading, reproducible snapshots, tracing, and long-horizon task budgets. The page links Coasty to Coarena, an open arena for computer-use agents, and names offerings such as environments, models and infrastructure, private evaluations, and custom benchmarks. The strongest published performance claim on the reviewed site is around OSWorld: Coasty says its own stack measured 85.6% and 82.8% independently verified on OSWorld. It also lists managed Windows and Linux VMs, a 150-step budget, and an 1,800-second run budget for a production model named `coasty-v5`. Those are homepage claims and should be treated as vendor-reported unless a buyer validates them through an independent evaluation. Coasty is best suited for AI labs, agent developers, and enterprises evaluating or deploying computer-use agents. It is not a no-code desktop macro tool. The product is aimed at teams that need reproducible task environments, private evals, custom benchmarks, and API-level access to computer-use workflows. Pricing is not published in the docs reviewed; the site points prospects to book a meeting, so production use should be treated as custom pricing. For an OpenTools page, Coasty fits the AI infrastructure category rather than the productivity-app category. Its value is the combination of task API, desktop environments, and evaluation methodology. Teams comparing computer-use agents can use it to reason about reproducibility, traces, and task outcomes instead of relying only on polished demo videos.