Skip to main content
AgentKit CLI commands are organized into groups; each group has its own page below, listing options and examples per command. Run agentkit tree any time to print the full command tree (--all expands args and options, --json for machine-readable output).

Global options

Place global options before the subcommand. They apply to the current command. The provider is resolved in this order: --provider, AGENTKIT_CLOUD_PROVIDER, CLOUD_PROVIDER, project configuration, user-level defaults, then volcengine.

init

Scaffold a new project from a template

config

Manage lifecycle config

build

Build a runtime image

deploy

Deploy an already built runtime

launch

Build and deploy a lifecycle project

status

Show runtime state

release

Run the full cloud release pipeline

runtime

Manage runtimes

destroy

Delete a deployed runtime

knowledge

Manage knowledge bases

memory

Manage memory collections

model-gateway

Manage Model Gateway, model providers, and consumers

mcp service

Manage MCP services

sandbox

Create sandboxes, run commands, manage sessions

skill

Manage skills on the skill platform

harness

Harness

chat

Chat with shared OAuth Harnesses through tenant aliases

invoke

Call a deployed runtime

eval backend

View evaluation backend

eval dataset

Manage evaluation datasets and cases

eval evaluator

Manage evaluators

eval target

Manage TEA evaluation targets

eval run

Submit evaluation experiments

eval experiment

View experiments and results

migrate

Migrate an existing agent application

auth

SSO authentication, profiles, and organization setup

onboard

Install bundled skills

tree

View the complete command tree

upgrade

Check for and install CLI updates

docs

View the docs in your browser
Last modified on September 19, 2026