fal Serverless

fal Serverless is an event-driven serverless deployment platform for running custom machine learning models, proprietary pipelines, and fine-tuned AI variants on auto-scaling GPU infrastructure. Developers write applications using the fal.App Python framework, defining model initialization logic (setup()) and endpoint routes (@fal.endpoint) that scale automatically based on incoming traffic.

The platform handles runner provisioning across GPU hardware (including NVIDIA H100, H200, and B200 instances), multi-layer caching for cold start reduction, and configurable concurrency controls (min_concurrency, max_concurrency, concurrency_buffer). Additional features include CLI deployment (fal deploy), versioned revisions with instant rollbacks, multi-environment support (staging/production), Prometheus metrics exports, and log drains. Billed per second of execution time.