Kalavai Developer Documentation

Welcome to the Kalavai Developer Docs — your guide to building, training, and deploying AI workloads on Kalavai’s distributed compute platform.

Kalavai leverages spare data center capacity to deliver flexible, cost-effective compute for machine learning, AI inference, and large model hosting.

What is the Kalavai Platform

Kalavai Platform is a managed computing platform that simplifies access to GPU compute and LLM hosting. It builds on our open-source orchestration library, integrating directly with tools you already use — like Ray and GPUStack — to provide on-demand distributed compute for AI workloads.

Efficient cost and low infrastructure overhead

Our platform abstracts the complexity of provisioning and managing GPU clusters, while optimizing performance and cost through dynamic utilization of spare capacity. When using the Kalavai Platform, users have direct access to a large fleet of data centre level GPUs at the lowest price in the market.

Product	Description
Managed GPU Clusters	Spin up distributed Ray and GPUStack clusters for training, hyperparameter tuning, reinforcement learning and custom workloads.
Async Inference & LLM queue	Affordable LLM inference for large scale intelligence projects
Model fine tuning	Easily customise LLM to your data

Beta Tester Program 🚀

We’re currently in Beta, and inviting developers and research teams to get early access to Kalavai. We're seeking developers who have hands-on experience with one or more of the following frameworks to participate in our beta testing program: Ray, Unsloth, Axolotl or GPUStack.

👉 Join the Beta Tester Program to get started.

Join our exclusive Discord community for beta testers.

Get started for free

Get a free account here to access the Kalavai Platform. All accounts come with access to free resources, like CPUs, GPUs and memory.

During the Beta Testing phase, you will be asked to join the Beta Program the first time you login. This will grant you free resources to test the platform.

If you need more resources, you can request more in the Clusters page, but note that capacity is limited.