For the complete documentation index, see llms.txt. Markdown versions of all docs pages are available by appending .md to any docs URL.

kagent init command

Initialize a new agent project with the specified framework and language.

kagent init [framework] [language] [agent-name] [flags]

Arguments:

• framework - The framework to use (currently supports adk)
• language - The programming language to use (supports python)
• agent-name - The name of the agent project

Flags:

• --description - Description for the agent
• --instruction-file - Path to file containing custom instructions for the root agent
• --model-name - Model name (e.g., gpt-4, claude-3-5-sonnet, gemini-2.0-flash) (default: “gemini-2.0-flash”)
• --model-provider - Model provider (OpenAI, Anthropic, Gemini) (default: “Gemini”)

Global Flags:

• --kagent-url - kagent URL (default: “http://localhost:8083”)
• --namespace, -n - Namespace (default: “kagent”)
• --output-format, -o - Output format (default: “table”)
• --timeout - Timeout duration (default: 300s)
• --verbose, -v - Verbose output

About kagent init

The kagent init command creates a bootstrap agent project. Select a specific model using the --model-provider and --model-name flags. You can also customize the root agent instructions by using the --instruction-file flag.

If no custom instruction file is provided, a default dice-rolling instruction is used. If no model is specified, the agent can be configured later.

Currently supported model providers are the models that are supported in the models configuration:

• OpenAI
• Anthropic
• AzureOpenAI
• Gemini

Example

Create a new agent project:

kagent init adk python dice

Create a new agent project with custom instructions:

kagent init adk python dice --instruction-file instructions.md

Create a new agent project with a specific model:

kagent init adk python dice --model-provider Gemini --model-name gemini-2.0-flash