Beam

Beam is a serverless GPU platform with sub-second cold starts and 1.7k+ GitHub stars — a Modal alternative for deploying AI inference, training, and code execution with automatic scaling.

Key features

Compute & infrastructure

• Sub-second cold starts with persistent GPU containers
• NVIDIA GPU support (T4, A10G, A100, H100)
• Automatic scaling from zero to thousands of containers
• Built-in object storage and volume mounts
• Custom Docker images and dependency management

AI workloads

• LLM inference with vLLM, TGI, and custom backends
• Fine-tuning and training with distributed GPU support
• Batch processing and async task queues
• Model serving with auto-scaling and traffic splitting

Developer experience

• Python-native SDK with decorator-based deployment
• Git-based deployments and CI/CD integration
• Real-time logs, metrics, and observability
• Team collaboration and role-based access control

At a glance

License	AGPL-3.0
Stack	Go, Python, Kubernetes
Self-hosted	Yes — self-hosted control plane
Cloud	Beam Cloud (managed)
GPU	T4, A10G, A100, H100

Self-hosting

git clone https://github.com/beam-cloud/beta9.git

Beam offers a self-hosted control plane option for enterprise deployments. The managed cloud version provides the simplest path to production.