Subcommands and options

The hai CLI has additional functionality that supplements the REPL.

Specify user

If you've logged into multiple hai user accounts, the default behavior is to use the last user account. You can force hai to run with a specific user account:

hai -u <username>

This requires that you've previously logged into the account with /account-login.

Specify model

To override the default LLM model for an invocation of hai, use -m:

hai -m <model>

This option is especially helpful when creating command aliases. For example, I use hai1 and hai2 to start hai with different models depending on what I'm doing.

# Use GPT‑5.1 Chat for general prompts
alias hai1='hai -m gpt-51-chat'

# Use Sonnet-4.5 for coding prompts
alias hai2='hai -m sonnet45'

Task mode

To immediately drop a user into task-mode:

$ hai task <task-fully-qualified-name>

To trust a task, use --trust.

Set API key for LLM provider

$ hai set-key <provider> <key>

Supported providers: openai, anthropic, google, deepseek, xai