How Kinesis works
From an app definition to a running workload, with control at each step.
Kinesis separates the app you want to run from the compute that runs it. You describe the container, runtime and resources; the platform coordinates deployment onto compatible capacity.
1. Define the workload
Choose an organization, then create an app inside a project you can edit. The organization owns the app and pays for its usage. Start with a container image or connect a GitHub repository for an image build. Configure environment variables, network ports, execution settings and resource requirements.
2. Choose where it runs
Use serverless CPU capacity for managed placement, or target a grid backed by dedicated cloud servers or infrastructure you bring. Available capacity, hardware requirements, placement constraints and grid quotas determine whether an app can be scheduled.
GPU apps require a grid. Serverless GPUs are currently paused.
3. Follow deployment
Creation, builds, provisioning and lifecycle changes can be asynchronous. An API response or successful portal action acknowledges the request. App status, build status and server orders tell you how execution is progressing.
4. Connect and operate
Use the configured endpoint for traffic, logs for application output, and metrics for resource usage. Configure the application's own authentication and network behavior to match its purpose. When an app fails to start, check its configuration and startup logs before retrying.
5. Review capacity and cost
Organization admins manage billing and infrastructure. Developers operate accessible apps; per-person grid quotas are charged against the app’s creator.
Stopping an app and releasing a dedicated server are separate actions. Check which resources remain allocated and their charging model before treating a workload as fully shut down.
The portal, REST API and MCP server provide different interfaces to these same resources.
Last updated
Was this helpful?
