OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Infrastructure & Serving

BETA9

BETA9 is an open-source runtime for running serverless AI workloads, with a Python interface for deploying autoscaling inference endpoints, isolated containers and background tasks on cloud or self-hosted infrastructure.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

BETA9 — OpenViST project profile in AI Infrastructure & Serving.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use BETA9 when you need runtime infrastructure to serve AI models, execute code in isolated containers, or queue background processing, and want a Python-facing workflow with cloud or self-hosted deployment options.

Who it is for

  • Developers deploying custom-model inference endpoints with GPU resources and autoscaling.
  • Teams needing isolated remote containers to run LLM-generated code.
  • Engineering teams processing asynchronous work through resilient task queues.
  • Operators choosing between self-hosting the runtime and using Beam’s managed cloud platform.

Documented uses

  • Deploy an autoscaling endpoint for a custom model and specify its compute resources.
  • Run code remotely inside an isolated container.
  • Define typed background-task inputs and a retry policy, then enqueue work.
  • Fan out workloads across parallel containers or scale serverless workloads to zero.
Before you choose

Requirements & limits

The README says BETA9 can be self-hosted for free or used through Beam’s managed cloud platform. Its quickstart calls for creating a Beam account. The repository license is AGPL-3.0.

  • The runtime supports both cloud GPUs and operator-provided GPUs.
  • An inference endpoint can use queue-depth autoscaling with configured container and task limits.
  • Background tasks can be invoked without deploying them, or deployed behind a versioned endpoint.
  • The README lists scale-to-zero, distributed volume mounts, webhooks, scheduled jobs and hot-reloading.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Define an autoscaling inference endpoint

Write a Python handler configured as an autoscaling inference endpoint with GPU resources.

Before you start

  • The beam-client package, installed as described in the README’s installation section.
  • Python 3.11 is specified for the endpoint image.
  • The example requests an A10G GPU, 2 CPUs, and 16 GiB of memory.

Steps

  1. Import Image, endpoint, and QueueDepthAutoscaler from beam.

  2. Decorate a handler with @endpoint, setting image=Image(python_version="python3.11"), gpu="A10G", cpu=2, and memory="16Gi".

  3. Configure QueueDepthAutoscaler(max_containers=5, tasks_per_container=30) on the decorator and define the handler’s return value.

Expected outcome

A Python handler is defined with endpoint configuration and an example response; the README describes this pattern as creating an autoscaling endpoint for a custom model.

Guide limitations

  • The README does not show how to deploy, invoke, or test this endpoint.
  • The example handler returns a fixed label and confidence value; model loading and inference are not demonstrated.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗Guide documentation ↗
Source record details
Source availability
Active
Sources checked
11 October 2026
Original website title
beam-cloud/beta9 · Official homepage title · retained in its source language
How we review sourcesSuggest a correction