Welcome
CosmicAC is a self-hosted platform for running GPU workloads on your Kubernetes cluster. It runs GPU Container Jobs and serves open source language models.
CosmicAC is a self-hosted platform for running GPU workloads. You deploy it on your host machine. It runs GPU Container Jobs and serves open source language models on your Kubernetes cluster.
CosmicAC job types
CosmicAC runs two kinds of job. Choose the one that fits your task.
Get started
Deploy CosmicAC, then install the CLI.
Once CosmicAC is running, create your first job.
Learn the concepts
Understand how CosmicAC runs jobs and serves models on your cluster.
Guides
Step-by-step guides for creating and working with jobs, and managing your deployment.
GPU Container Job
Create a GPU Container Job, open a shell into it, and set up SSH over Tailscale.
Managed Inference Job
Create a Managed Inference Job, create an API key, and call the model it serves.
Platform management
Upgrade your deployment, manage racks and model masters, and set up notifications and observability.
Reference
Look up the commands, routes, fields, and values CosmicAC accepts, and what changed in each release.
CLI commands
Every CosmicAC CLI command, with usage, arguments, and options.
Task deployment commands
Every task command that deploys, upgrades, and operates the stack.
API reference
The inference, monitor, and observability settings HTTP routes.
Configuration reference
Deployment environment variables, kubeconfig requirements, and job fields.
Recommended model parameters
Recommended serving parameters and hardware for supported models.
Changelog
New features, changes, and fixes in each release.