Skip to main content

Why Terminal for agents

Agent workloads are not one-size-fits-all. An interactive coding step may need fast tokens now, while a background review, eval, or research job can wait if that makes it cheaper. Terminal is designed around a simple idea: every agent step should define its own latency and cost SLA. Coming soon, you will be able to set those requirements directly in request headers without changing endpoints. Today, use the Standard window for interactive work and Flex when a job can be patient. See the planned header format. You only pay per token. Terminal is serverless by default, with no deployments to manage and no rate limits to size around.

Base URL

Endpoint

Authentication

Terminal is currently invite-based. Contact the Terminal team to request access. All examples in these docs use the placeholder TERMINAL_API_KEY.

First request

Current models

Additional serverless OSS models are available by request. See Models supported.

Service windows

Next steps