River API

River API is a developer platform for sampling and fine tuning hosted AI models through a single Python client. It provides programmatic access to model inference and training workflows, including supervised fine tuning, LoRA based adaptation, reinforcement learning, checkpoints, and long running asynchronous jobs on managed GPU infrastructure. Developers authenticate with a River API key and can sample from the models available to their account through the river client package. Training sessions can create LoRA adapters, run supervised learning steps, optimise models with reinforcement learning objectives, save checkpoints, resume work later, and sample from trained weights. Many operations run asynchronously on GPU workers, while the high level client can wait for results automatically. River API is differentiated by combining hosted inference with model training in one developer workflow. The documentation lists multiple large open models, but model access is account scoped and the catalog can change. Usage depends on compute and model access, so buyers should check live console pricing, enabled models, rate limits, training requirements, and data handling before moving production workloads onto the platform. For FutureStack, River API belongs in Developer Tools because its primary buyer intent is building and training custom AI systems. It is best for developers and ML engineers who need hosted inference, fine tuning, reinforcement learning, and checkpoint based experimentation without managing the entire GPU stack themselves.