Runtimes and model accounts
Choose the runtime each agent runs on, connect model provider accounts, pick models, and check that a runtime works before you rely on it.
A runtime (in the API, an adapter) is the engine an agent runs on: Claude, Codex, Gemini, Cursor, OpenCode and others. Each runtime signs in to a model provider with an account and uses a model you choose.
Which runtimes are available
Your organization's settings list the runtimes it may use. Organization administrators can turn runtimes on or off; agents can only use runtimes that are on.
Model accounts
A model account is the credential a runtime signs in with: a provider sign-in or an API key. Connect one while hiring an agent or from Connectors.
- One account can serve several agents.
- An account belongs to the organization, or to a person when it is a personal subscription.
- Keys are stored as encrypted secrets.
Choose a model
On an agent's Runtime tab, choose the model. The list shows the models the connected account can use. Saving checks that the account still has access; it does not start a run.
Test a runtime
Select Test on the agent's Runtime tab. Veyra starts the runtime in the agent's environment and checks that it can authenticate and reach the model. A passing test means the agent can work; whether a task succeeds is shown by its run.
Subscriptions and API keys in costs
Runs on an API key record their per-token cost. Runs on a subscription (for example a provider plan) record the subscription separately, so Costs can show the fixed fee apart from usage.
From the API
See Agent runtimes and AI connections in the API reference.