deploy deploys the runtime described by the current project’s lifecycle config. It does not build an image first; use launch when you want build and deploy in one command. Use release for the .agentkit/agentkit.yaml cloud release pipeline, IM channels, or frontend release flow.
Run agentkit build in the project first so the same lifecycle configuration contains a build result. Cloud and hybrid modes require runtime-management permissions in the target provider and region; local mode requires Docker
Deployment creates or updates the selected runtime, can replace a serving application, and may incur cloud charges. Check the provider, region, and runtime name first
deploy
Read agentkit.yaml, select the local, cloud, or hybrid strategy from common.launch_type, and write deployment state such as runtime id, endpoint, or local container state back to the config file.
deploy runs service preflight checks before deployment. In an interactive terminal it prompts when required services are not enabled; use launch --preflight-mode when an automated build-and-deploy flow must control preflight behavior explicitly.
Harness Deploy
With --harness <name>, the CLI reads <name>.harness.json from the current directory, converts it into runtime environment variables, creates a temporary lifecycle config, then builds and deploys in the cloud. If you pass --discovery-url, also pass at least one allowed client id through --allowed-id.
Harness deploy creates or updates a cloud runtime and may rewrite harness.json in the current directory with endpoint, runtime id, and auth information. Confirm the working directory, Harness name, and target project before running it.
Runtime roles
For runtime creation, an explicit IAM role must exist. Otherwise the CLI reuses a role with AgentKitDefaultRuntimeAccess or creates a new one. Updates retain the existing role without modifying its policies automatically; scheduled tasks configure their required TOS prefix permissions separately
After deployment, inspect state with agentkit status and verify an application response with agentkit invoke run "hello". The first checks runtime state; the second invokes the application and model. For model authentication failures, check runtime environment variables before rebuilding
