Headless and one-shot runs

For a single answer without the UI:

belai -prompt "what does internal/run do?"
belai -provider anthropic -model claude-sonnet-4-5 -prompt "review this diff"
belai -plan -mode plan -prompt "plan the migration to the new API"

-prompt sends one turn, prints the reply and exits. Everything else works as in the TUI: the same classifier, permission rules, guardrails and budgets.

Flags for scripts

Flag	Meaning
-prompt TEXT	One turn, then exit
-mode	agent, plan, goal or code; left out, the mode is classified from the prompt (the TUI’s auto)
-plan	Start in read-only plan mode
-provider, -model, -effort	The model
-agent NAME	Run an agent profile by name in the foreground
-tools	Tool execution, on by default; -tools=false turns it off
-answers FILE	A JSON file of answers, given in order to the questions the run asks, since nobody is there to type them
-usage-json PATH	Write a JSON summary of token usage to this path on exit
-no-transcript	Keep no session transcript
-detect-mode	Print the mode the classifier would pick, and exit
-no-git-sync	Do not sync the branch before turns
-trust-dir	Trust the directory first
-firewall	Route model traffic through the active AI Firewall

With nobody at the terminal, an ask that cannot be answered is refused unless you allow it: -allow-ask-without-tty and the other -allow-* flags loosen specific gates on purpose. CLI reference lists every flag.

Resuming and exporting

Flag	Meaning
-resume ID (or -r)	Resume a session by id or prefix
-continue (or -c)	Continue the most recent session here
-teleport ID	Continue a session from another host. See Teleport.
-export ID	Write a session as Markdown

Related

• Sessions
• CLI reference